Digital Identity in Cyberspace
White Paper Submitted for 6.805/Law of Cyberspace: Social Protocols
Professors Hal Abelson and Lawrence Lessig
10 December 1998
Paul Covell (email@example.com)
Steve Gordon (firstname.lastname@example.org)
Alex Hochberger (email@example.com)
James Kovacs (firstname.lastname@example.org)
Raffi Krikorian (email@example.com)
Melanie Schneck (firstname.lastname@example.org)
Table of Contents
I. Overview: What is Digital Identity?*
A. Working Definition of Identity*
B. Identity as Commodity*
C. Verifying Versus Revealing An Identity*
II.Introduction to Cyberspace *
A. The Premise*
B. The Promise of Unbundling*
C. Privacy: Type I Unbundling*
D. Anonymity: Type II Unbundling*
E. The Technology*
F. The Promise of Unbundling Revisited*
IV.Anonymity vs. Accountability *
B. The Internet*
C. Link and No-Link: An Architectural Choice*
D. No Link*
E. Link Architecture*
F. Preventing Crimes*
V.Mandatory Authentication Mechanism - Constitutionality *
B. Traceability as a precondition for speech*
C. Traceability as a precondition for access*
D. Constitutionality of Traceability as a Precondition for Access*
E. Fourth Amendment Analysis*
1. Introduction: The Driver’s License Analogy*
2. Introduction to Fourth Amendment Doctrine*
3. Constitutionality of Mandatory Traceability Under Current Fourth Amendment Jurisprudence*
F. Applying the Reasonableness Test*
1. Constitutionality of Traceability Under Potential Modifications to the Reasonableness Test*
G. First Amendment: The right to anonymity*
1. Mandatory Traceability Under the Compelled Speech Doctrine*
B. Digital Certificates*
C. Using Digital Certificates*
D. Digital Certificates and the Unbundling Paradigm*
E. Securing Digital Certificates*
F. Anonymous Certificates*
G. Using Digital Certificates for Traceable Anonymity*
H. Certificate Authorities*
I. Completing the Picture*
VII.Business Aspects of Digital Identity *
B. A Note on Architectural Choice*
C. The Business-Consumer Relationship*
1. The Ability to Authenticate Information*
2. The Ability to Unbundle*
D. The Business-Business Relationship*
A. Community in cyberspace is based on the interaction between people*
B. Examples of Community*
C. Anonymity Today*
D. Implications of Full Unlinking*
E. Implications of Mandatory Linking*
IX.Road to Implementation *
A. From Here to There: The Road to Implementation*
B. Social Norms Barriers*
C. Market Barriers*
D. Legal Barriers*
E. Architectural Barriers*
Appendix 1: Legal Process For Law Enforcement Access To The Trace*
Appendix 2: Digital Signature Legislation*
A. State Digital and/or Electronic Signature Legislation*
1. Prescriptive Model*
2. Criteria-based model*
3. Signature-enabling legislation*
B. National Conference of Commissioners on Uniform State Law (NCCUSL) Draft Laws*
1. NCCUSL Efforts: The Uniform Electronic Transactions Act (UETA)*
2. Scope of the UETA*
3. Provisions of the UETA*
C. Pending Federal Digital and/or Electronic Signature Legislation (see Table)*
Table of Figures
Figure 1: Degree of Privacy*
Figure 2: Degree of Anonymity*
Figure 3: Possible implementation of proposed architecture*
Figure 4: Unbundling in the Dynamic of the Marketplace*
Currently there is no generic system for identification in cyberspace. It is not possible to absolutely identify an entity or to accurately tell whether an object has a specific characteristic. Digital environments have inherent differences from real space which causes this discrepancy, and when implementing an identity system for cyberspace one needs to consider more than just the architectural nature of the system — any system chosen will have social repercussions which need to be also taken into account.
Identity is a unique piece of information associated with an entity. Identity itself is simply a collection of characteristics which are either inherent or are assigned by another. The color of a person's hair and whether or not another thinks he is attractive is part of a person's identity.
Interactions done in real space inherently carry the identity of the person originating the transaction. Generally, physical traits are carried along in a transaction — for example when one purchases a book from a book store, the book dealer may remember the buyer's face or build.
The difference between real space and cyberspace is that the essence of any digital transaction is unbundling. Ones and zeros do not inherently carry any separate information along with them; a real space transaction carries along inseparable secondary information. Digital transmissions can only transmit; there is no secondary information encoded in the transmission unless explicitly put there. Thus, for authentication purposes, additional information needs to be carried with cyberspace transactions for identity purposes.
Providing extra information in digital communication introduces the possibility for identity theft. Since nothing prevents the transmission of false identity information, or the duplication of another's identity information. To prevent these problems, the actual identity must not be transmitted along with the message; instead a verification scheme needs to be used to convince the recipient that the message was actually sent by the sender. This eliminates the need to send one's actual identity. The concept of verifying instead of revealing provides an extra layer of security to the sender.
The other point of insecurity is in the digital certificates were issued to verify these characteristics. These certificates are meant to be used only by their owner, but if they are obtained by another party, then that party can falsify his identity, representing himself as the individual for whom he has digital certificates.
Architecturally, we must decide how to store and use these certificates. The certificates can be stored on a smart card for use on a computer terminal, or the certificates can be stored in an "identity server" locked via password or biometric information and available for transmission over the Internet.
In real space, it is difficult to selectively to verify or reveal portions of one's identity: most forms of identification contain more information than is needed for any transaction. The unbundling that is possible in cyberspace allows portions of identity to be disassociated and verified by a third party. This not only creates the ability to verify via the least revealing means, but it also creates the framework for anonymous transactions — it is possible to merely verify the proper information without ever distributing the name characteristic. Further, cyberspace users have control over the strength of the link between their real world and cyber-identities.
That is, in cyberspace, users can unbundle identity from content and transactions.
Therefore, designers of an identity system for the digital environment need to consider whether or not to build a system that facilitates traceability — i.e. whether or not to build a system in which it is always possible to trace one's cyberspace content/transactions to one's real world identity. This question is extremely important as it will not only affect the architectural design of the system, but it will also have side effects disturbing governmental, commercial, and social environments in cyberspace.
Forcing a mandatory link to identity (i.e. mandating traceability) provides properly authorized law enforcement with a crucial tool in criminal investigations — the ability to determine with whom criminals are interacting. Law enforcement also will have the ability to monitor illegal activities and trade, and easily determine who is involved. However, commerce may also leave the United States if other nations provide anonymous transactions. The mandatory identity link will also stifle dissenting speech as citizens will be afraid to voice their opinions because everything they say can be traced to their identities.
Not providing the link is detrimental to law enforcement since they will have no means to track crime in cyberspace (especially once encryption becomes more widespread and law enforcement lose the ability to get the content of any transmission). Commercial and social interests have been fulfilled if there is no traceability, except now both will have to coexist in an environment in which there will be a significant amount of criminal activity.
Another issue that needs to be considered by those designing a digital identity system is related to the ability to separate characteristics from identity. This may create a market for personal characteristics. A person may now have the ability to sell a personal characteristic to another party in exchange for goods or services. It may also be possible for companies to collaborate and share people's characteristics in order to recreate as much of the identity as possible for marketing purposes. Situations such as these need to be fully analyzed and appreciated before the design of a digital identity system can begin.
Any decision to unbundle characteristics in the creation of a digital identity system cannot be made simply by choosing an architecture that is simpler or more elegant to implement — there must be a consideration of the ramifications this decision is going to have on the cyberspace community.
This paper explores in detail the legal, political, social and technical issues surrounding the development and adoption of an internet architecture that permits individuals to exchange authenticated information. Before proceeding with that discussion, however, it is important to examine the concept of identity itself. This section develops a working definition of identity, considers the ways in which people use their identities, and articulates the reasons why it is important to protect our identities, especially in the digital context.
It is difficult to craft a formal definition of identity. Basically, the essential and unique characteristics of an entity are what identify it. These characteristics might include, among other things, the unchanging physical traits of the person, his preferences, or other people’s perceptions of the individual’s personality. The skills that a person possesses can also become part of one’s identity. For example, a person’s identity could include the fact that he "has the ability to drive" or that he "has brown hair." Some characteristics, such as height, have one correct setting. Those traits of an individual that reflect someone else’s perceptions do not have to have an absolute setting. Bob may set Alice’s "is friendly" flag to true, whereas Charles may set the same flag to false. Even if Bob and Charles agree on what should be the flag’s setting for Alice, Alice’s own view may differ from theirs. Thus, in practice there is a degree of fuzziness to the definition of an entity’s identity, and most certainly to how it is perceived by others.
No two identities are the same. Each identity maps to a unique set of characteristics. Two people may share some of the same characteristics, such as being old enough to drive or having the same hair color, but that does not mean that they have the same identity. One simply is not looking at enough characteristics. Upon further inspection, it may be found that one of the individuals has brown eyes and the other blue eyes. Therefore, when someone perceives that two identities as the same, he should search for new information that adds details that distinguish the identities from each other. Although this section speaks of identity as enabling the ability to distinguish individual people, it should be noted that identity can be used to discern individual corporations or fictional characters as well.
Identity also evolves over time, with more characteristics becoming evident everyday. When someone purchases groceries, one of the characteristics that may be added to that person's list of characteristics is that he "bought bottled water from the supermarket." As a child gets older more facets of his personality may become apparent. Other characteristics may change their state. A simple example would be a change in hair color.
One may vary the way in which one represents the characteristics that make up an identity. One way to represent these characteristics to create a binary flag, consisting of either a "yes" or "no," that states whether the person possesses the desired characteristic. Although it is possible to define every characteristic as a binary flag, it may not always be appropriate. Creating binary flags that say whether an individual lives in each nation of the world -- "yes" if true, "no" if false -- may not be as efficient in answering the question, "Where do you live," as creating a single flag that represents the nation in which the individual does live, say for example, the "United States." Different uses of identity call for different representations.
The distinction between characteristics and identity is not firm. Often, a unique characteristic serves as the representation or identifier for identity. Consider social security numbers. Each social security number is unique and can be used to identify an individual , but no social security number itself contains all of the characteristics included in that person’s identity. In practical terms, full names serve the same function for most people.
So what uses or purposes do our identities, and the characteristics from which they are constructed, serve? For one, identity allows someone to address an individual without confusing him with others. Identity functions as a cue that allows us to access our memories for information on someone.
Identity also is used in commerce when we purchase items from retailers or sign documents, for example. In part, the needs of the transaction will determine the amount of one’s identity that needs to be used. Some transactions do not rely on the ability uniquely to identify an individual. For example, at the liquor store, all that is required is some proof that you are of legal age to purchase alcohol. The store only needs to know whether you have the characteristic that includes you in the class of people eligible to purchase alcohol. Other transactions depend on the unique identification of the individual -- which requires knowledge of an identifier -- to work, but do not require knowledge of the individual’s full identity. For example, for a postcard to reach your mother it will need to be labeled with her mailing address. Finally, some situations may require full knowlege of an entity’s identity. To an extent, a successful marriage may require each spouse to know the full identity of the other. Similarly, a successful lawyer-client relationship may require rather complete knowledge of an entity’s identity, at least with respect to a particular issue.
In today’s economy, identity information often is viewed as a valuable commodity. This view of identity is worth a closer examination.
Businesses desire to advertise their products to the markets most interested in them, and may even retool their products to be more appealing to certain segments of a market. Knowing the preferences of individuals allows a corporation to target perfectly their products to those who would prefer and, thus, be most likely to purchase, them. Making a detailed survey of an individual’s preferences, though, is very difficult, if not impossible. Often an individual cannot specify the exact motivation for her purchase of a particular product. From the seller’s perspective, determining which questions to ask purchasers can be a daunting task. Further, certain questions, despite their potential usefulness, are not likely to be answered by a purchaser. To work around this problem, businesses use identity information as a proxy for preferences. For example, rather than trying to discover the exact reason why an individual purchased a Ford Mustang, a car dealer might instead try to find out the purchaser’s profession or income level. Suppose the car dealer discovers that a number of his customers who have purchased Ford Mustangs are lawyers. Although the car dealer may not understand why they purchased Ford Mustangs, he can assume with some level of confidence that there is something about lawyers that leads them to purchase Mustangs instead of Cougars.
Other products have clear markets. The obvious market for judicial robes, for example, is members of the judiciary. In this case, identity information serves as the means by which a business can determine who is part of the market. In the abstract, all businesses use identity information in this manner. One clear market for every product is "paying customers." Knowing that an individual is "credit-worthy" or a possessor of cash helps businesses identify the members of this market.
As a result, many businesses collect information about identity as part of its transactions. A purchase order form may ask for an address, occupation, and income level. Stores may ask individuals to relinquish portions of their identity in exchange for goods and services. For example, customers may be offered special discounts or free products if they complete a survey. Similarly, a customer may be asked to complete a registration form detailing her reading and television-viewing habits in order to receive a card entitling the customer to a discount . The business creates a database using the information from these forms and the names of the products that were purchased. Hoping to develop a more accurate profile of their customers -- in essence, hoping to learn the full identity of the average consumer, businesses sell or rent portions of their databases to other businesses. Conceivably, if enough vendors collaborate, a "profile" of buyers may be created without the consumers express permission or knowledge. The information is then used to guide the direct marketing of other products to customers or the retooling of current products. It also could be used to identify those people who have a high probability of not paying their bills. Such databases stand to threaten the privacy interests of consumers, especially for those purchasing legal, but socially stigmatized, products like pornography.
In addition to leveraging transactions in the marketplace, our identities serve other purposes. For example, he formation and preservation of our identities, as a collection of our characteristics and traits, are important for our psychological and emotional well-being. Throughout history, mankind has struggled with the essential questions of life: Who am I? What am I supposed to do? Who am I supposed to be? All of these questions are tied closely to how we view our identity and how we construct it. For some individuals, seeing their identity used as the tool of commerce -- important because it helps a business identify how to make a buck from you -- is damaging to their psychological health. Other people are psychologically bothered by the idea that someone can find out information about them without having to get all of the information directly and explicitly from them. Depending on the power imbalance between the parties, the commodity nature of identity may exacerbate the feelings of lack of control, importance, and purpose about which many individuals of modern society complain that they suffer.
Cyberspace creates opportunities for identity theft. One inherent property of digital media is that it can be duplicated perfectly and easily. Exact copies of everything sent over a digital communications channel can be recorded.Consider the act of sending a signed letter to someone. In real space, I reveal to the recipient the exact form of my signature, but the difficulty of mastering the art of forgery protects me from the possibility that the recipient would begin signing letters with my signature. However, if I send a digital letter that contains the digital representation of my signature, the recipient could easily duplicate and use my signature to assume my identity when signing documents. The seriousness of this problem is highlighted when you consider that future technologies will allow extremely important identifiers, such as a retinal scan or a fingerprint, to be represented digitally. These biometric characteristics are protected in real space because they are embedded in the physical body of the person. This is lost in cyberspace.Thus, cyberspace needs a system that allows individuals to verify their identities to others without revealing to them the digital representation of their identites. A verification system would let Bob, for example, know the identity of Alice or that she possesses a particular trait, but would not give him the ability to impersonate Alice or use the trait identifier as if it was his own. In our digital letter example, Bob would be able to verify that the letter contains Alice’s signature but would not let him sign documents as Alice. Similarly, a verification that someone is of the proper age to purchase alcohol would not give the person verifying this identifier anything that would allow him to represent himself as being of the proper age to purchase alcohol. Such a sytem helps both parties obtain what they want out of exchanging identity information without the risk of identity theft.
Cyberspace is "the total interconnectedness of human beings through computers and telecommunication without regard to physical geography." Cyberspace is "a term coined by science fiction author William Gibson to describe the whole range of information resources available through computer networks." For our purposes, cyberspace is a realm in which communication and interaction between two individuals or between an individual and a computer is facilitated by digital data exchanged over computer networks. This interaction or communication can be used for a host of different purposes.
The Internet is currently the biggest network for linking computers, but cyberspace as a concept is independent of the Internet. Cyberspace communication began before the Internet and the World Wide Web, and cyberspace interaction and communication will continue to take place after the Internet is no longer the network of choice.
Two general metaphors are often used to explain and define cyberspace. In the first, cyberspace is viewed as a geographic "place" to which one can go. Much of the end-user terminology relating to cyberspace is based on this metaphor. For example, one can "visit" a Web site and "enter a chat room." Even the company name "Netscape" highlights this land-based, geographic metaphor.
The second metaphor focuses on communication, viewing cyberspace as a conduit for information. This view emphasizes the actual network technology rather than the community aspects of cyberspace. Despite its multitude of cables, routers , and switches, and the wide variety of applications layered upon it, fundamentally the Internet, is a means by which one computer communicates with another.
Neither of these metaphors perfectly encapsulates cyberspace, and both are necessary to understand fully the range of issues that are raised by interactions in cyberspace. The geographic metaphor is limited because in fact at each end there is still a person sitting somewhere in real space—people are not disembodied just by participating in cyberspace. However, the geographic metaphor is useful because interactions in cyberspace are more than just transitory voices on the telephone—there can be a lasting digital record of the communication, and people can return later in time to continue the same conversation (e.g., through postings in an on-line discussion group). Because cyberspace interactions can be saved and continued over time, cyberspace communities can develop. However, the communications metaphor lends a certain amount of realism to a complete understanding of cyberspace. Because there is a person sitting in real space on each end of the communication, at some level cyberspace communication is just a "souped-up" phone call, and at this level the same concerns and rules that apply to analog voice telecommunications could apply to cyberspace. However, exchanges can be much more complex and far more efficient in cyberspace. These differences in degree ultimately may amount to differences in kind.
Cyberspace can facilitate an enormous range of uses. The two most common applications currently are e-mail and Web browsing. Both of these are flexible tools that can be used for almost any purpose. In the social sphere, cyberspace can enable communication between two specific individuals. Or, one individual can publish information on-line for the general public to access. Businesses can, in addition to exploiting these communications methods, use cyberspace for transactions among businesses or transactions between businesses and consumers. In some cases, the entire transaction can be completed in cyberspace; in other situations, some elements of the transaction must occur in real space. Additionally, cyberspace can, and perhaps must, be used not only for business and social purposes, but for regulatory purposes (e.g., taxation and law enforcement) as well.
A critical problem in cyberspace is knowing with whom you are interacting. In essence, the problem is that "on the Internet, nobody knows you’re a dog." Currently, you cannot determine accurately the identity of the person on the other side of an e-mail message or know with certainty the source of any information in cyberspace.
(The New Yorker, Vol. 69 (LXIX) no. 20, page 61, July 5, 1993, by Peter Steiner )
A digital identity system must serve several functions. First, authentication—ensuring that when a message purports to be from Alice, Alice sent it, not someone pretending to be Alice. Second, message integrity—providing certainty that when a message arrives from Alice, it is the same message that Alice sent, not modified en route in any way. Third, non-repudiation—ensuring the inability of Alice later to deny that she sent the message, and the inability of the recipient of Alice’s message to deny that the message was received. Finally, establishing a digital identity architecture may have the beneficial side effect of facilitating confidentiality through encryption—the knowledge that no one besides Alice can read a message intended for her. For our analysis in this paper, a digital identity system must serve the first three functions, and may serve the fourth.
The technical problem with cyberspace in 1998 is that there is no effective, widespread architecture to verify identity on the Internet. There is no digital identity mechanism that meets the needs for the diverse range of cyberspace interactions. While there are currently systems that attempt to solve this problem, they all fail for various reasons. Some fail because they are not secure or reliable. Some fail because they only work in very narrow contexts and are not interoperable. Some fail because they attempt to apply literal translations of real world identity and do not seek to capture the benefits of making identity digital, such as the unbundling of traits, which will be analyzed below. Some fail because they mandate very rigid rules and do not allow flexibility around the policy trade-offs involved in any identity verification system. Finally, some fail to be adopted because they are too expensive.
E-mail addresses are currently the most widespread form of digital identity in cyberspace. People use an e-mail address as an identifier because e-mail is the most direct and easy way to reach a person in cyberspace. However, the current e-mail architecture has little security and includes no reliable identity verification. The dominant protocol for sending e-mail (SMTP) does not facilitate verification of the sender’s identity, and therefore does not facilitate authentication: an e-mail message may purport to be from "email@example.com" there is no certainty that Bill Gates actually sent it. It is a trivial technical task to forge the source of an e-mail message under the current architecture. Likewise, e-mail is not safe from tampering en route and can be repudiated after it is sent or received.
In many cyberspace contexts, passwords are used to verify a person’s identity. However, passwords are easily shared or distributed. Providing a correct password proves only that the user has knowledge of the password, not that the user really is any particular person. There is no certainty that after issuance a password remains only with its intended holder and has not been distributed through innocent or malevolent means. There is thus no secure link between a password and any particular real world or cyberspace identity. Nonetheless, passwords are easy to implement and perhaps are better than using no security measures (although they may give false confidence), so they have become by far the most widespread method for cyberspace identity ‘verification.’
In the e-commerce realm, credit card numbers often are used as a form of identification. However, this is an inappropriate form of identification. It is over-revealing, and it is not intended to identify any personal traits. A credit card number is only a payment mechanism. However, many e-commerce vendors, particular those selling digital pornography, use credit card numbers as a proxy for proof that an on-line user is over 18 years old. This is not a correct assumption, as those under 18 legally can have credit cards. In addition, providing a credit card number gives the vendor much more information about the user than the mere fact that the user may be over 18 (and the ability to charge the user an appropriate fee). With a credit card number and access to consumer credit reporting databases, a vendor could find out enormous amounts of personal information about the user based on this one identifying number.
Internet Protocol (IP) addresses serve as the fundamental roadmap of the Internet. IP addresses allow data to reach the correct computer on the Internet. In this capacity, these numbers are critically important. However, as an identity mechanism for Internet users, they are severely lacking. First, IP addresses link only to a computer and do not help in any way to identify the person who is using the computer. Second, many computers are now connected to the Internet with dynamic IP addresses (vs. static addresses), meaning that each time the computer connects to the Internet it uses a new and temporary IP number. This makes tracking identity—even of computers—based on IP numbers impossible.
The current identity architecture of cyberspace is thus in great need of an overhaul. In order to facilitate flexible but secure verification of digital identity, a new cyberspace identity infrastructure is needed.
One major premise of our project is that new technology facilitates unbundling of identity information and therefore has the potential to provide a degree of control over privacy and anonymity in cyberspace that is difficult, if at all possible, to achieve in the real world. This section briefly explains this premise.
The concept of unbundling captures two distinct notions, each of which facilitates the exercise of choice. First, to unbundle identity is to treat identity as a set of individual traits, rather than one integrated bundle of traits. Second, when identity is unbundled, a set of traits need not be bound to a single "real world" person. For clarity, we refer to these types of unbundling as Type I and Type II unbundling, respectively, discussing each type of unbundling in turn.
The ability to achieve Type I unbundling facilitates control over one’s degree of privacy. When identity is unbundled, it can be treated as a set of individual traits rather than one integrated bundle of traits. Identity is bundled perfectly when all of one’s traits are grouped together. Identity is unbundled partially when small groups of individual traits are packaged together. Identity is unbundled completely when it is divided into a set of individual traits treated separately. When identity is partially or completely unbundled, entities may be identified by one or a combination of traits, rather than by a complete set of traits. Thus, Type I unbundling facilitates control over the degree to which one’s identity is revealed. Indeed, it facilitates choice regarding which and how many elements of identity to reveal.
Control with respect to revelation of identity (i.e., the ability to choose which and how many elements of identity to reveal) consequently facilitates apparent anonymity, full disclosure, and selective revelation of identity. At one extreme is apparent anonymity: no elements of identity are revealed. At the other extreme is complete identification: all verifiable elements of identity are revealed. These extremes mark the endpoints of a spectrum of choice representing varying degrees of privacy. As we progress along the spectrum from apparent anonymity to complete identification, we selectively may reveal more and more elements of our identity (see Figure 1). Such selective revelation of identity may occur on an ad hoc basis, or may be based on defined principles.
There are several principles in accordance with which one selectively might reveal elements of identity. One such principle is the "least revealing means" principle. In accordance with this principle, one would choose the least revealing means of identification necessary to serve a purpose (e.g., the least revealing means necessary to complete a particular transaction). Under an alternative principle, the "most convenient means" principle, one might choose to reveal more information. For example, in order to enable a software agent to find a product to match one’s preferences, one might be willing to reveal more information than dictated by the "least revealing means" analysis. Under the "most convenient means" principle, one selectively would reveal the combination of identifying information that would provide the most convenience provided that there is an upper bound beyond which one would not be willing to reveal information for convenience.
By enabling us credibly to assert traits about ourselves, digital certificates facilitate the Type I unbundling that provides a degree of control over privacy. Indeed, we have seen how Type 1 unbundling facilitates apparent anonymity, complete identification, and selective revelation of identifying information because it enables us to treat identity as a set of individual traits rather than one integrated bundle of traits. The general concept of unbundling identity also includes Type II unbundling, to which we now turn.
Type II unbundling of identity refers to the ability to control the strength of the link between our cyberspace and real world identities and to the corollary that a set of traits need not be bound to a single "real world" person. The ability to achieve Type II unbundling facilitates control over one’s degree of anonymity. Indeed, with Type II unbundling, we can have a strong link, weak link, or no link at all between our cyberspace and our real world identities. Where there is a strong link between these identities, we are the person that we represent ourselves to be in cyberspace. However, as the link between our cyberspace and real world identities becomes more tenuous, the malleability of our cyberspace identity increases. Taken to its extreme, unbundling enables us to cut the link between real world and cyberspace identity, leaving us with two completely different identities: one in the real world and one in cyberspace (See Figure 2). Thus, true anonymity is possible in cyberspace.
Further, once a set of traits has an existence independent from a real world person, there is no bar to multiple identities. Not only might we have different identities in the real world and in cyberspace; we also might have multiple cyber-identities with different, and even conflicting, traits.
The key point is that new technology facilitates unbundling of identity information. The new technology to which I refer is digital communication, and, more precisely, the digital certificate. For purposes of this section of the paper, it is sufficient to note that digital certificates enable us to make credible assertions about ourselves in cyberspace. Certificates can be used to verify a trait or to verify actual identity. For example, I can present my "I am over 18" certificate to verify that I am eligible to vote, and I can present my "I am Melanie" certificate to verify my identity. The interesting point is that digital certificates enable us to make credible assertions both about our traits and about our identities, and thereby facilitate the Type I and Type II unbundling that provide a degree of control over privacy and anonymity.
In the real world, there were a number of obstacles to unbundling. However, in cyberspace, it possible to unbundle identity to a degree heretofore quite difficult, if at all possible, to achieve in the real world. With the technical ability to unbundle, we theoretically are able to choose any degree of privacy or anonymity represented along the two spectra, including perfect privacy and perfect anonymity.
As we develop a model within which to think about the design and development of an architecture for digital identity, we must keep in mind the flexibility of the technology with which we are working, particularly because seemingly innocuous design features may mask social choices of profound significance. As Langdon Winner wrote:
In our times, people are often willing to make drastic changes in the way they live to accord with technological innovation; at the same time, they would resist similar kinds of changes justified on political grounds. If for no other reason than that, it is important for us to achieve a clearer view of these matters than has been our habit so far.
Political theorists have long been in the business of trying to form the perfect society: one in which the people are content and prosperous and there is no crime. Approaches to developing this society have varied, from the Hobbes’ tyrannical world to Marx’s non-governmental communist scheme. However, in all cases a central issue is order: keeping people within the society from harming one another, or preventing crime. In order to prevent crime, society creates a regulatory body designed to enforce laws. Law enforcement functions by providing disincentives to breaking the law and a system of procedures to deal with those who do. Law enforcement protects people from criminals and criminal activity; however, almost every political philosopher (excepting Hobbes) has recognized the interest society has in protecting itself from law enforcement. That is, law enforcement must be given a certain amount of power in order to prevent people from committing crimes and to punish those who do commit crimes; however, because law enforcement itself must be composed of members of society, an attempt must be made to avoid the abuse of power. This is the classic conflict between liberty and order.
In the United States, the First and Fourth Amendments attempt to provide this protection. The First Amendment limits the extent to which the government can regulate speech, and the Fourth Amendment protects citizens from unreasonable search and seizure. This creates a tension within society: society has both a need to protect itself from its individual members who may transgress against its rules, and simultaneously a need to protect itself from the very same law enforcement body which it created to solve the initial problem. This tension effectively sets the landscape for any discussion of the rights of law enforcement and citizens within society: law enforcement will push to gain as many privileges as possible so as to guarantee their effectiveness in fighting crime, whereas society will fight to limit law enforcement's power to only those things which it deems critical to performing its duties. This line between "critical" and "ideal" will vary with time and with circumstance, and must therefore be reexamined when the needs of society change.
It is important to note that in the real world, anonymity and accountability are interrelated in such a way that the liberty to speak with perfect anonymity may come at the cost of accountability, or order. In cyberspace, however, we will see that we need not provide anonymity at the expense of accountability. As total unbundling becomes a technical possibility in cyberspace (i.e., in the absence of friction that prevents unbundling in the real world), we must, as a society, strike an acceptable balance between anonymity and accountability; between liberty and order. As we attempt to strike this balance, it is important to resist the temptation simply to mimic the real world in cyberspace; instead we must consider the underlying values of privacy, anonymity, and accountability as we construct a digital identity system.
The introduction of the Internet as a widely accessible medium for commercial, social, and government interactions has created the necessity to reexamine the needs of society. With its introduction comes an unprecedented ability to transfer information extremely quickly to a large number of recipients at extremely low cost. Unfortunately, this ability is a double-edged sword: although it enhances the ability of the law-abiding public to participate in commerce and communication, it likewise enhances the ability of criminals to do the same. However, there are some ways in which the architecture of cyberspace is fundamentally different from the architecture of the physical world. These differences force a reevaluation of the rights and responsibilities of law enforcement within this new world.
Scott Charney and Kent Alexander outline the several ways in which computers can be involved in crime. In all cases a harm is committed which would be considered a crime in the real world, except that a computer is involved in some way. The scenarios involve computers A and B. In the first case, data on computer A is stolen, erased, or damaged. In the second case, computer B is used to commit a crime: this can be either a traditional crime, or a crime involving a victim computer, "A," as in the first case. The third case is one in which a computer is not a target or source of a crime, but contains evidence that a crime was committed, or was used in planning the crime.
Clearly, all of these cases cause harms that should be prevented or punished by law enforcement. The real world contains a certain amount of "friction" which can be used to identify the perpetrator of a crime and expose the extent of the harm. A police officer can question a suspicious person regarding their identity or intentions, and ultimately detain or arrest that person if sufficient cause is present. People leave physical evidence or traces of crimes: fingerprints, hair, carelessly forgotten articles of clothing, or a description of their physical appearance. With cyberspace comes the potential to eliminate the real world friction, which creates traceability. Information transfer, whether legal or illegal, could be rendered completely anonymous and untraceable.
Current Internet technology facilitates some of this friction. While it may be cumbersome, it generally is possible to determine the real life identity of someone who commits a crime on the Internet. However, as in the physical world, this friction is an artifact of the architecture of the Internet. More importantly, it is an artifact that could be eliminated: truly untraceable communication could be facilitated for the first time in history. However, as evidenced above, law enforcement has strong interests in preserving traceability. A complete response to the question of whether or not to allow completely untraceable information will require the involvement of the affected society; making an informed decision means understanding the implicit choices that are concomitant with the larger choice. Once the implicit choices are understood, the choice to include link within the architecture can be made based upon the societal values embodied within each choice.
As identified earlier, any digital identification system must determine where to lie upon the continuum of anonymity and accountability; that is, a system must adopt an appropriate degree of Type II unbundling. However, within the context of law enforcement it becomes clear that not all points along this continuum are equal. One point is very different from all the others: the point at the far end of the spectrum where there is absolutely no traceability. For the sake of clarity in future discussion, this point will be called "no-link." At the no-link point, there exists within the digital identification architecture no mechanism for determining the link between data in cyberspace and the real world recipient or sender. The no- link point implies only that there is no mandatory link between cyberspace and the real world; this does not preclude an additional, non-mandatory method of determining identity that could be layered on top of the no-link architecture. All other points along the spectrum will be designated as "link" points. This indicates that there is some mandatory architectural mechanism for determining the real world identity of the sender and receiver of data in cyberspace.
Is this really an architectural choice? Perhaps legal, social, or market constraints will provide a more suitable method than technological constraints for regulating the presence of link on the Internet. They would certainly provide a more flexible system than one that technologically required a real world link to be maintained for every data transfer. The Internet requires an "open" architecture, one that different entities can customize to a large degree in order to obtain the effect they want for their users. This is one of the reasons the Internet is already so vast and will continue to be viable: to a large extent people have the ability to define how users will interact with their site. Regulations of a legal, social, or market nature will be subordinate to this general feature of Internet architecture, and that subordination will place a strong restriction on the effectiveness of regulation as a universally applicable tool.
Architecture can thus be seen as enabling communication, with content providers placing restrictions on user interaction. For example, a matchmaking service may require that all users provide a real world name and phone number for contact information. In this way the architecture enables any such site to place restrictions on the ways in which users may interact with the site. Users who do not wish to conform to the restrictions presented by a particular site can decide to utilize a site that more closely matches their preferences.
Sites sharing many common identification requirements can be collected together and identified as a "domain." Domains may be as broad or as specific as necessary, but three major domains are likely to dominate interactions on the Internet: business, social, and government domains. The business and social domains both have interesting conceptual questions associated with them, but for the most part they are simply a translation of business in the real world to business in the cyber world. Not much is fundamentally different: contracts are still contracts, markets are still driven by supply and demand, and advertising is still fundamental to success. The situation is similar for the social domain. While novel methods of interaction may be created and offered for idea exchange, the fundamentals of social interaction remain the same in cyberspace. Since the business and social domains largely translate, their cyberspace identification requirements are likely to be similar to their real space identification requirements.
This leaves the government domain, which breaks down into two major areas of operation. First, the government domain includes service providing sections, such as the Social Security Administration and the IRS, which are responsible for providing government services to the public. In this capacity, the government functions very much like a business, imposing its own set of restrictions upon user interaction; the main difference is that in some situations users will not have the choice to simply choose not to participate. Discussions of universal access are beyond the scope of this paper, but could potentially have a large impact when dealing with government interactions in cyberspace. The other major area of government operation consists of regulatory powers. Specifically, the duty of governments, from municipal to federal, to provide law enforcement.
If we adopt a no-link system, then government would need to form a domain of its own facilitating traceability in order to facilitate law enforcement. However, this is impossible: in order for law enforcement to adequately perform its duties, its domain must impact all other domains. For example, a method of determining a specific user's real life identity would be effectively useless if it were not available in all domains in which a crime could potentially be committed. There is a fundamental difference between all other domains and the domain of government regulatory powers. It is a domain that cannot be adequately provided for within the open architecture framework, because law enforcement capabilities must not be dictated by the domain in which the user is interacting.
Ultimately, then, this link/no-link decision must be made at an architectural level. While conceivably laws, market, or social norms could be designed to create a link situation, each would only be effective against those who participate within the legal, market, or social system: it would be ineffective against criminals, who would have little regard for such influences and feel little responsibility or pressure to obey their restrictions.
Both link and no-link architecture have benefits and drawbacks associated with them. With a link architecture, access to the link information can be limited, presumably, only to an appropriately regulated law enforcement agency with specific regulatory processes in place for obtaining the information. However, the immediate point is that not everyone will have access to the information contained in the architectural link; to those without access, a link architecture is identical to a no-link architecture. The benefit of identification is still present, but the ability to gain knowledge of the person's real world identity from the architecture of the system is limited to those specific bodies with access. Thus, once again, the interesting area of discussion is that pertaining to law enforcement: when can a link system effectively be used as a no-link system, and are there benefits to being able to determine link which outweigh any corresponding drawbacks?
At all points along the continuum, except for the extreme of one-to-one identity, there is a need to distinguish between "transient anonymity" and "persistent anonymity." With transient anonymity, no persistent link remains to the sender of the information; this is analogous to anonymous leafleting. Persistent anonymity is perhaps more useful: it allows continuity of cyber identity generally without disclosing real world identity. It only permits disclosure of the real world identity within a link system. In a no-link system, continuity is preserved, but without facilitating link. Both types of anonymity are useful in some circumstances, but persistent anonymity is likely to be more generally useful.
The benefits of a no-link system are, as mentioned above, those pertaining mostly to issues of freedom of speech and freedom of action. In the commercial domain, the wheels of capitalism are greased by the no-link architecture. People who have no fear of ever being personally associated with what they buy are far less likely to be concerned about the social norms which might have previously restricted them from purchasing a product. Unbundling facilitates the necessary degree of identification that commerce will require without necessitating the revelation of the entire real world identity. Free speech is likewise assisted by the absence of traceability: where potential oppressors are unable to determine the sender's real world identity, there is no danger of oppression.
To provide certain services, the government may require knowledge of certain aspects of real world identity. However, in other circumstances the government has no need to know identity. For example, in the voting context, government can utilize the unbundling potential of identity in cyberspace to determine a person's right to vote and their proper district, without knowing the person’s real identity.
The benefits to be derived from absence of real world identity in social and commercial interactions are realizable in the context of political dissident and otherwise unpopular speech and purchases. Ultimately this end of the discussion becomes a question of how much individuals will trust their governments to only reveal identity when absolutely necessary for authorized law enforcement purposes. "If only popular ideas were protected, we wouldn't need a First Amendment." If there is no fear of unreasonable retribution from the government, viewpoints which would have previously been considered unpopular and socially unacceptable will become far more visible; J.S. Mill argues that this sort of "marketplace of ideas" is absolutely necessary to the ability of society to make well reasoned judgements.
In a no-link architecture, individuals do not have to trust (or fear) the government at all. In the United States, this issue might be less of a factor than in other countries; however, due to the global nature of the Internet, as well as the lapses of judgment that can occur in any governing body, the importance of this issue must not be understated. Should Internet users in other countries have the same rights of free speech that we in the United States have deemed valuable? This raises additional logistical and sovereignty issues, and may even threaten the global nature of the Internet, if countries decide not to participate in a system that facilitates complete untraceability.
No-link architecture provides protection from McCarthyism. But in so doing it removes all accountability from speech. It is an architecture that completely eliminates the power of social norms, market regulation, and legal regulation to govern interaction on the Internet. Society should not overlook the more general consequences that may result from the ability to avoid accountability in all speech, especially speech which would not be considered criminal: people may routinely and without concern spout inaccurate and misleading information, and responsibility may disappear even further from the moral landscape. However, the aspects which can be most clearly identified and discussed are those which result in criminal behavior.
As mentioned earlier, the computer can be involved in crime in several different ways. As an anonymous method of communication, the potential for the Internet itself to become involved in crime becomes very high. Private communication can be utilized as a method of planning crimes, and public communication groups can become breeding grounds for criminal activity. Charney and Alexander mention that "it might be possible to allow individuals to congregate in certain places where anonymity is assured, with each individual participant on notice as to the benefits and risks associated with anonymous communication." However, this seems foolhardy: each individual may fully understand the implications of participating in an unrestricted, anonymous area, and yet the potential hazards to society in terms of the planning of criminal activity would continue unchecked. Additionally, crimes that can and should be prosecuted in the real world involving restricted speech, such as libel and child pornography, can occur unrestricted in an area with no-link.
The issue then becomes one of preventing crime, while simultaneously attempting to mitigate this potential "chilling effect" on free speech. At the heart of this discussion lies the distinction between transactional information and content information. Transactional information is information regarding the sender, recipient, and other information associated with the transmission of the information but not regarding the content of the information. Thus far the argument has centered around transactional information; however, the value of content to law enforcement must be considered: if it is absolutely necessary to have content as well as transactional information, then it will do no good to consider offering the latter without the former. If, on the other hand, transactional information without content is a tool that can be utilized, it may represent an effective compromise between the needs of law enforcement and the desires of society.
Encryption represents the single largest barrier to law enforcement obtaining content from a computer. This is an issue that is relatively unique to cyberspace, as handwritten and telephone encryption is relatively rare. One choice can be made with respect to encryption: allow it, without regulation, or disallow it. Disallowing encryption altogether is pragmatically different from allowing only key escrowed encryption, but for the purposes of this discussion they are effectively the same. The overwhelming response of government has been that encryption controls are in fact necessary, and several initiatives have been proposed to this effect; however, both the public and legal reaction to these initiatives has been negative: many organizations are resisting the degree of control which law enforcement would be given, and the Communications Decency Act was recently ruled as too general to be constitutional. In this situation, law enforcement's claims of what it needs to be effective are strongly disputed by the public: the equilibrium between the two is harder to strike in cyberspace.
While most encryption can no longer be broken by brute force methods, the alternative procedural approach to gaining access to content is to subpoena the key necessary to decrypt the content. This is satisfactory because it allows traditional warrant restrictions to apply in the area of gaining access to content. At the same time, however, it is hopelessly flawed because a clever (or even computer illiterate) criminal could easily "misplace" the key, or intentionally delete it upon receipt of the subpoena.
This would tend to argue for the necessity of government controlled encryption. The main drawback to this argument is a practical one mirroring the gun control argument: when encryption is outlawed, only outlaws will have encryption. While the argument is slightly more complicated than this, the basic point remains that encryption is too ubiquitous and inexpensive to expect sophisticated criminals to use a "government approved" version. Likewise the social and market opposition to this sort of key escrow system is large and entrenched, rendering encryption control a pragmatically infeasible solution.
Given the choice between unregulated encryption with link, and unregulated encryption without link, law enforcement will almost certainly choose to have the possibility of determining link. Without it, the case is fairly strong that it will be extremely difficult to prevent criminal activity. The constitutional issues surrounding such a requirement on all speech will be examined in Section V. Regardless of the legal status, however, if a link architecture is decided upon, there will be chilling effects on free speech, and all the benefits of a public forum for interaction may be lost.
The negative implications of choosing the link system are clear: it may place an unreasonable burden on free speech. Even if it is not unconstitutional in this manner, it may simply deter people from speaking out in situations where their voices would be most useful. In order to convince society that its interests in avoiding unreasonable persecution are maintained, the architectural decision to include link must be combined with legal regulations regarding who is given sanction to disclose the link, and under what circumstances such disclosure is acceptable. While the negative impacts of providing a link with all transmitted data can never be fully accounted for, the goal of a system which provides an architectural link must be to mitigate the impact of the architecture as fully as possible.
A no-link architecture has more tangible drawbacks. Crimes can be easily planned and carried out on a system with no accountability, and there is no reason to think that they wouldn't be. However, practical concerns such as sovereignty and providing unrestricted speech to political dissidents regardless of their government's policy on free speech may outweigh the potential societal costs. It may be also that suitable mechanisms for regulating identity can be created in a legal or market based way; it is hard to see how these methods would be enforceable in a cost-effective manner, but the number of criminal deviants might be small enough that identification by law enforcement could be reasonably achieved.
The ultimate question is, to what extent is law enforcement empowered to track down information, and to what extent does that empowerment place law abiding citizens at risk? This question must be combined with the immediate concern that a lack of empowerment will cause criminal activity to propagate unchecked. Perhaps the current system of cooperative tracing provides enough of this "friction" to allow the real world methods to be comfortably adopted in cyberspace; however, it seems that the ongoing march of technology will dictate other solutions, ones which rely on intent and not on friction to bring about the desired results. In order to determine adequately what the intent should be, it will be necessary to examine carefully the implications of constructing a system with no friction so that the intent is accurately reflected in the results.
Traceability refers to the existence of a link between a cyberspace identity and a real world identity. Mandatory traceability refers to the requirement that such a link be created and maintained. Mandatory traceability generally takes one of two forms. First, to achieve mandatory traceability, one might require identification as a precondition of speech in cyberspace. Alternatively, one might require identification as a precondition of access to cyberspace. We briefly consider and reject the former and then consider the constitutionality of the latter form of traceability in detail.
One way to accomplish traceability is to require identification as a precondition of speech (e.g., by outlawing the transmission of anonymous messages from a particular access provider, from a class of providers, or in a network). This facial form of traceability would require a user openly to attach his identity to his message, and, as one commentator points out, only would withstand strict scrutiny in cases in which the identity requirement implicated "other overriding constitutional rights, such as voting rights or property rights." When this type of identification requirement is designed simply to combat the risks of fraud and corruption on the electoral process, it will not survive constitutional scrutiny. Further, such a broad ban on anonymous speech is unlikely to survive; indeed, an identification requirement in this form would bar the use of services such as anonymous re-mailers, and likely would be struck down. We do not propose a system of digital identification in which traceability is a precondition for speech because the technology available permits us to craft a more narrow identification requirement which better would serve the state’s legitimate law enforcement ends.
Identification could be required as a precondition for access to cyberspace using law, technology, or a combination thereof. Using law, the government potentially could regulate access to cyberspace whether access is obtained through government-subsidized or private ISPs. In the case of private ISPs, the government could require the ISPs not only to require identification as a precondition to access, but also to keep logs of cyberspace users linking their cyber-aliases to their real world identities. Further, the government could provide ways in which legal process could be used to compel private ISPs to respond to authorized law enforcement requests for identity information.
Alternatively, traceability could be implemented at the infrastructure level. That is, a digital identity architecture could be inserted into the existing network infrastructure so that users cannot access a networked information system until they identify themselves. Thus, properly authorized law enforcement officials will be able to trace a user’s activities to his identity.
A more detailed discussion of the law and technology on which our proposed implementation relies can be found in Sections VI and IX of the paper, however the basic idea behind any system of mandatory traceability is that speakers entering cyberspace would be required to deposit (e.g., with the ISP), or attach to their communications, a means of tracing their identities. One can conceptualize mandatory traceability by positing a regime in which an encrypted fingerprint automatically would be attached to every transaction in cyberspace. In such a regime, the fingerprint could be encrypted with the government’s public key such that properly authorized law enforcement officials could access the private key necessary for decryption while participants in the cyber-transaction would not be able to strip away the speaker’s anonymity.
We now consider the constitutionality of requiring traceability as a precondition for access to cyberspace. In order to allow an individual to maintain full control over privacy and anonymity in the cyber-environment, yet ensure the availability of real world accountability in limited circumstances, we imagine the implementation of a flexible digital identity system conjoined with a system of mandatory traceability. This section will grapple with the constitutionality of such a digital identity system, focusing on the First Amendment right to anonymity and the Fourth Amendment right to privacy and protection against unreasonable searches and seizures. For the remainder of this section, we will use the designator "system of mandatory traceability" to refer to a fully flexible digital identity system that simply creates and maintains a link between one’s real and cyber-identities.
Before entering into our doctrinal analysis of mandatory traceability under the Fourth Amendment, we would like to identify an analogy that is useful in thinking about the constitutionality of a traceability requirement. Consider the driver’s license: the small, plastic card that people carry around in real space in order to signify that they are qualified to operate a motor vehicle. One’s driver’s license contains information that identifies the holder of the license: name; birthdate; sex; height; weight; identification number; and picture. In addition, the card has verification features that demonstrate that the information on the card has been certified by an authorized, trusted authority, the DMV. For example, in order to verify the validity of the license, it is laminated, it may have holograms that are difficult to forge, and it may include signatures or other indicia that the state attests to the truth of the information on the card.
What is the primary purpose of a driver’s license? Most people likely would say that a driver’s license verifies that the holder of the license is qualified to drive. However, a license whose purpose simply was to indicate that the holder was qualified to drive, could be comprised of much less information than currently is included on the license. In particular, a license seeking to serve this purpose could be comprised of a trait (a statement that the holder of the license is qualified to drive), a link from the trait to the individual with the trait (a picture of the qualified driver), and a means of verifying the validity of the link between the trait and the individual with the trait (the DMV’s signature, or other means of verifying that the person pictured has proven that he is qualified to drive).
What purpose does each piece of information serve? The trait ("the holder of this license is qualified to drive") tells us the purpose of the license. The picture ties the trait to the holder of the license. It ensures that the ability to drive, noted on the license, is attributable to the actual holder of the license, and therefore prevents individuals from transferring their credentials to others. Finally, the verification vouches for the validity of the link between the trait and the person pictured on the license. Note that as this example demonstrates, a license need not include the driver’s name in order for the license to serve its designated purpose. Yet, our licenses carry our names.
What purpose does it serve to have our name on a license when a picture clearly is sufficient to demonstrate that the person presenting the license is qualified to drive? Why does the state require that our names be on our license? Further, why are we required to carry our license when we drive—why is it a violation to drive without a license? Why have I committed a violation when, as a qualified driver, I forget my license at home? If one is qualified to drive, why should one be required to carry his license? That is, why must the law-abiding citizen carry his driver’s license? I submit that the requirement that I carry my license is designed to ensure that drivers can be held accountable.
Indeed, the name on our licenses, in combination with the requirement that we carry a license, ensures accountability. These requirements ensure that we can be held accountable for traffic violations as well as traffic accidents. I am required to carry my name with me when I drive so that my identity can be ascertained by law enforcement officials in the event that I commit a traffic violation, or am involved in an accident. Thus, we introduce the example of the driver’s license to raise a point—the driver’s license is a perfect example of an identification requirement that is imposed, before one has caused harm, in order to preserve accountability in the event that one causes harm. Requiring that drivers carry a license with their name on it ensures that they can be held accountable for their actions.
With this in mind, we now turn to our doctrinal analysis of mandatory traceability under the Fourth Amendment. Those familiar with the Fourth Amendment may wish to go directly to the analysis of the constitutionality of mandatory traceability.
The Fourth Amendment to the United States Constitution provides: "The right of the people to be secure in their persons, houses, papers, and effects, against unreasonable searches and seizures, shall not be violated, and no Warrants shall issue, but upon probable cause, supported by Oath or affirmation, and particularly describing the place to be searched, and the persons or things to be seized." The plain words of the Amendment indicate that searches and seizures are subject to a "reasonableness" requirement, and that probable cause is required for a warrant to search or seize. In interpreting the amendment, courts constantly have struggled both to determine what constitutes a search subject to the warrant requirement, and to interpret "reasonableness."
As the Court struggled to define the scope of the Fourth Amendment, it needed to determine just what constituted a search. In the early twentieth century, the Supreme Court’s Fourth Amendment jurisprudence was geared toward the protection of property. The Court’s inclination to protect property quite clearly is reflected in its 1928 decision in Olmstead v. United States (277 U.S. 438 (1928)). In Olmstead, the Supreme Court held that use of a wiretap to intercept a private telephone conversation was not a "search" for purposes of the Fourth Amendment. One of the grounds on which the Court justified its result was that there had been no physical intrusion into the person’s home. Under Olmstead’s narrow view of the Fourth Amendment, the amendment was not applicable in the absence of physical intrusion. Thus, without trespass or seizure of any material object, surveillance was beyond the scope of the Fourth Amendment as interpreted by the Olmstead Court.
However, in its well-known decision in Katz v. United States, 389 U.S. 347 (1967), the Supreme Court rejected Olmstead’s "trespass" doctrine, articulating, in its place, a Fourth Amendment jurisprudence based on the protection of individual privacy. In Katz, the Court held that the Fourth Amendment protects people, not places: "What a person knowingly exposes to the public, even in his own home or office, is not a subject of Fourth Amendment protection¼ But what he seeks to preserve as private, even in an area accessible to the public, may be constitutionally protected." Thus, the Court held that physical penetration of a constitutionally protected area is not necessary before a search and seizure can be held to violate the Fourth Amendment. According to the Court in Katz, "once it is recognized that the Fourth Amendment protects people—and not simply "areas"—against unreasonable searches and seizures it becomes clear that the reach of that Amendment cannot turn upon the presence or absence of a physical intrusion into any given enclosure." Thus, although the Government’s activities in Katz involved no physical intrusion, they were found to have violated the privacy on which the petitioner justifiably relied and thus constituted "search and seizure" within the meaning of the 4th Amendment. Changing technology precipitated the shift from protection of property to protection of privacy, and in 1968, just one year after Katz, Congress passed Title III of the Omnibus Crime Control and Safe Streets Act authorizing microphone surveillance or wiretapping for law enforcement purposes, and requiring a warrant, based on probable cause, prior to such surveillance or wiretapping.
The Court has struggled not only to define the scope of Fourth Amendment searches and seizures, but also to apply the "probable cause" requirement. Originally, it was assumed that "probable cause" was required for every law enforcement activity that constituted search or seizure. Subsequently, on suspicion short of probable cause, the Court permitted certain searches and seizures on the basis that they were reasonable—that law enforcement interests warranted a limited intrusion on the personal security of the suspect.
There are a variety of circumstances in which the legitimate law enforcement interest in searching may outweigh the invasion that the search entails. First, where there is great, imminent public danger, a less demanding standard than probable cause might be appropriate. Second, where there is rapidly disappearing evidence or where there are rapidly disappearing suspects, we might want to accept less than probable cause. Finally, we might require less than probable cause where the intrusion occasioned by the search or seizure is limited. Thus, there are a variety of circumstances under which the legitimate law enforcement interest in searching may outweigh the invasion that the search entails, thereby occasioning the application of a less demanding standard than probable cause.
The Court first departed from its rigid application of the probable cause requirement in Camara v. Municipal Court, a case involving the inspection of dwellings for housing code violations. Noting that even one undetected safety code violation could cause "fires and epidemics [that] ravage large urban areas," the Court applied a balancing test that involved "balancing the need to search against the invasion which the search entails." Thus, Camara marked the first time that the Court recognized that some Fourth Amendment activity should be judged under a balancing test.
One year after Camara, building on the balancing approach described in that case, the Court handed down its famous decision in Terry v. Ohio. In Terry, the Court upheld the power of police to "stop and frisk" suspicious persons without meeting the demanding standard of probable cause. More precisely, the court held that the constitutionality of a "stop and frisk"—the law enforcement practice of briefly detaining suspicious persons on the street for purposes of investigation—is governed not by the warrant requirement (i.e., probable cause), but "by the Fourth Amendment’s general proscription against unreasonable searches and seizures." In Terry, the Court tailored the level of suspicion required to the intrusiveness of the search, "opening the way for a sliding scale in which the less intrusive the search, the less demanding the procedural requirements for the search to be ‘reasonable.’"
In a system with mandatory traceability, the government will be able, under certain limited circumstances, to obtain an individual’s real world identity. If the user’s identification information is encrypted for protection, the government will require the decryption key. Eliciting the decryption key, which an individual seeks to preserve as private, likely would be considered a seizure in accordance with the privacy-based rationale articulated in Katz. Similarly, even if the user’s identification information is not encrypted, but simply is stored by a certificate authority or other trusted authority, the government may be conducting a search/seizure subject to the strictures of the Fourth Amendment when it attempts to elicit the identity information. Therefore, we must determine whether or not such searches are constitutional under the reasonableness test.
One way to ensure that a search passes the reasonableness test is to condition such a search on the presence of a warrant based on a showing of probable cause. Where a decryption key (if required) and a person’s identity will be revealed only in cases of individualized suspicion (i.e., cases in which there is a valid warrant), the search clearly would be deemed reasonable under our current jurisprudence. Indeed, one’s identity would be revealed only in those cases in which law enforcement has the right to access that identity and the person has no right to conceal his identity.
Even in the absence of a warrant, such a search potentially could pass constitutional muster. When authorized law enforcement officials obtain a decryption key so that they can decrypt the identity of an individual, and even when they obtain the person’s identity, the intrusion is minimal—there is little or no collateral damage of the sort with which the Court has shown concern. The government does not need to enter an individual’s private space in order to access the identity information, and the government will have access only to the identity information and to nothing else. Indeed, given the tiny burden on any particular individual’s life, traceability might be constitutional under the reasonableness test even with a lesser showing than probable cause.
Of course, this entire analysis presumes that there are sufficient assurances that the key and any identity information obtained would not be misused. Toward this end, the government might impose limitations on the circumstances under which law enforcement may obtain and use identity information, or other procedural limitations that would prevent misuse of revealed identity information.
While the traceability requirement likely would pass the Fourth Amendment’s "reasonableness" test, this test has come under criticism from some scholars. For example, Michael Adler argues that the current "reasonableness" test does not protect important interests that were protected under the old "property-based" standard, and therefore suggests that we modify the test to include these interests. We now consider two proposed modifications to the "reasonableness" test based on two distinct policy interests.
First, Adler suggests that we adopt an autonomy-based rule—a rule that focuses on an individual’s desire for control over his or her personal expression. In particular, he suggests a bright-line rule in which searches of the home or office need to be based on "individualized suspicion [that] would require the government to assemble first a reasonable belief based on information already outside the control of the individual. In other words, not until the individual has acted with the understanding that there could be telltale traces outside of her zone of control, thereby knowingly risking public attention, does she become vulnerable to government intrusion." Arguing in a similar vein, Scott Sundby suggests that "[w]hen factual probable cause is the core regulating device of government behavior, the [Fourth] Amendment is basically self-regulating because control over the government’s ability to intrude rests primarily with the individual. So long as a person does not engage in behavior arising to probable cause . . .individual privacy cannot be invaded." Under this rationale, mandatory traceability would survive constitutional scrutiny because the government’s ability to obtain one’s identity is limited to those situations in which the government has demonstrated an appropriate degree of individualized suspicion based on the individual’s activity.
A second modification of the Fourth Amendment’s "reasonableness" test relies on a substantive interpretation of the Fourth Amendment. According to a substantive interpretation, the Fourth Amendment was used to delineate the scope of government’s substantive power. In a 1995 article, The Substantive Origins of Criminal Procedure, William Stuntz suggests that historically, the privacy protections afforded by the Fourth Amendment were really "a proxy for something else, a tool with which courts or juries could limit the government’s substantive power." In particular, Stuntz argues that the Fourth Amendment’s focus "seems to have been to make it harder to prosecute objectionable crimes—heresy, sedition, or unpopular trade offenses in the seventeenth and eighteenth centuries, regulatory offenses in the late nineteenth century." Building on this interpretation of the Fourth Amendment, Michael Adler suggests that in the past, "the protection from arbitrary searches provided an unacknowledged but potentially quite important pocket of privacy in which individuals might be free to resist the state’s demands." Indeed, Adler argues that we ought to preserve spaces in which the government’s power to enforce the law is limited because such spaces allow for a degree of criminal activity that is necessary in order for society to flourish. The underlying notion is that perfect government enforcement of criminal laws would prevent the civil disobedience which often "provides society an impetus to reevaluate the law." Adler’s proposed modification of the "reasonableness" test would "preserv[e] the possibility of a low level of criminal activity and of allowing individuals some freedom from the punitive power of the state" by preserving spaces in which the government’s enforcement power is limited; spaces that serve as a source of power, independent of the government, that can be used to meet the collective need for a potential for disobedience. However, a system of mandatory traceability likely would pass the modified Fourth Amendment test because mandatory traceability does not enable law enforcement officials to enter spaces that it previously had no power to enter; it simply enables law enforcement officials who have witnessed a crime (or potential crime) to identify the perpetrators when the officials have made a proper showing of cause. Thus, mandatory traceability likely will survive scrutiny under the Fourth Amendment’s reasonableness requirement and the potential modifications thereof. We turn now to a First Amendment analysis of a system implementing mandatory traceability.
Although the Supreme is best know for its protection of free speech, it also protects rights ancillary to free speech including the right not to speak. The constitutionality of a system of mandatory traceability depends upon the scope of the limited right to anonymity that the Supreme Court has carved out for First Amendment protection.
Doctrinally, the right not to speak can be thought of as part of the compelled speech doctrine of First Amendment law. One scholar suggests that the compelled speech doctrine is comprised of three distinct prongs: confidentiality, autonomy, and preventing conscription. The confidentiality prong of the compelled speech doctrine, which protects the right not to speak, will be our focal point.
When properly implemented, a digital authentication system that facilitates traceability will permit the government to access identity information only in those circumstances in which the government has shown proper cause. Under the proposed system of mandatory traceability, the public will not be able to link a person’s real and cyberspace identities; only authorized law enforcement officials will be able to do this, and only in accordance with proper procedures.
The proposed traceability requirement attempts to preserve the ability to speak anonymously on the Internet. Although traceability would be required for Internet access, one’s speech on the Internet would appear to be anonymous. Thus, while law enforcement would benefit from the ability, under carefully circumscribed conditions, to access the identity of speakers, the speaker would remain anonymous with respect to the population to whom he or she addressed thoughts. Since the traceability requirement allows speakers to retain their anonymity at all times with respect to their audience, and to retain their anonymity with respect to the government at all times except in the narrow circumstances in which government officials are authorized to learn their identity, such a requirement might be limited enough to survive McIntyre. Indeed, the traceability requirement under consideration is a much more limited identification requirement than that in McIntyre which essentially would have sacrificed the anonymity of political speakers.
The Supreme Court in NAACP v. Alabama ex rel. Patterson, 357 U.S. 449 (1958) held unconstitutional Alabama’s demand that the NAACP reveal the names and addresses of all of its Alabama members and agents. The NAACP court is said to have recognized that "[s]erious First Amendment questions arise . . . when there is such a nexus between anonymity and speech that a bar on the first is tantamount to a prohibition on the second." As one district court explained: "[t]he Court in NAACP v. Alabama was of the opinion that the injury to a right subsequent to disclosure of identity precludes the right to identification." In NAACP, the Court recognized freedom of association and held that forcing the NAACP to divulge its membership lists was "likely to affect adversely the ability of [the NAACP] to pursue their collective effort to foster beliefs which they admittedly have the right to advocate." Thus, anonymity was deemed necessary to the exercise of freedom of association.
Unlike the public disclosure of NAACP membership lists, the disclosure of identity at issue here would not result in public dissemination of an individual’s identity, but would only consist of disclosure to authorized law enforcement officials. If properly implemented, a traceability requirement need not hinder one’s ability to exercise one’s freedoms, including the freedom of association. Indeed, when properly implemented, a system of mandatory traceability will not trigger the NAACP concern because the identification requirement will not subsequently restrain an individual in the exercise of an independent right.
Underlying NAACP was a concern that an identity requirement would raise a "fear of reprisal [that] might deter perfectly peaceful discussions of public matters of importance." In the past, registration laws have been overturned for infringing on the freedom of association only "when a history of harassment and social hostility is proven." In NAACP, for example, the Court noted:
On past occasions revelation of the identity of [the NAACP's] rank-and-file members has exposed these members to economic reprisal, loss of employment, threat of physical coercion, and other manifestations of public hostility. Under these circumstances, we think it apparent that compelled disclosure of [NAACP’s] Alabama membership is likely to affect adversely the ability of petitioner and its members to pursue their collective effort to foster beliefs which they admittedly have the right to advocate, in that it may induce members to withdraw from the Association and dissuade others from joining it because of fear of exposure of their beliefs shown through their associations and of the consequences of this exposure.
Of course, one might argue that a system implementing traceability would awaken a fear of reprisal and deter an individual from the exercise of his constitutional rights. However, such a claim constitutes a mere assertion, and NAACP is "inapposite where . . any serious infringement on First Amendment rights. . .is highly speculative." Further, where proper procedures are adopted to ensure that the government will not misuse its power to trace, it will not be realistic to argue that one’s speech will be deterred by a requirement that one’s cyberactivities be traceable to one’s identity in those limited circumstances in which law enforcement officials have shown proper cause.
Thus, even under the strict scrutiny applied to cases in which the government restricts political speech, it is likely that the traceability requirement will pass constitutional muster because it is narrowly tailored. In particular, under a system implementing traceability, one maintains one’s anonymity with respect to everyone except for properly authorized government officials, and anonymity need not be sacrificed to these officials save in limited circumstances. Further, properly authorized government officials only will be able to access one’s identity, and not other information about one’s activities. Thus, mandatory traceability is narrowly tailored to meet a substantial state interest—the state interest both in deterring crime in cyberspace and in holding accountable those who have committed crime in cyberspace.
Under the current interpretation of the First and Fourth Amendments to the Constitution, a traceability requirement likely would be deemed constitutional. In particular, we propose a flexible system of digital identity in conjunction with a mechanism that implements traceability in which a warrant is required for access to identity information. A warrant requirement would ensure that our system will meet the standards of the Fourth Amendment, and even in the absence of a warrant requirement, the trace likely would be held "reasonable." In addition, the ability to maintain the full spectrum of privacy and anonymity with respect to all but properly authorized law enforcement officials should enable a mechanism for traceability to survive scrutiny under the First Amendment.
This section provides an overview of the technical issues concerning the development of the digital identity architecture proposed in this paper. The section also explores a possible method for implementing the traceability feature that law enforcement might desire in a digital identity architecture.
The cornerstone of any digital identity scheme is a method for authenticating people and messages. It is important to be able to authenticate people, confirming characteristics about them such as their name, age, citizenship, or other relevant details. It is also important to verify messages, confirming that the person or server from which you are supposed to be receiving information is actually the entity sending the information.
The Internet is a large, open, public network. Unencrypted messages are sent in clear text, and are easily intercepted. Open Internet standards for e-mail, like SMTP (Simple Mail Transfer Protocol), have no authentication and forging an e-mail from another person is a trivial task. Web exchanges, through HTTP (Hypertext Transfer Protocol), also transmit data openly and can be easily intercepted. Internet standards were created to exchange information, not protect privacy and commercial transactions. To guarantee that the information your software receives is from the entity that should have sent it, there is a need for digital authentication.
There are two basic paradigms for establishing digital authentication. One is a trusted system, where information is presumed to be correct and you trust the computers involved and security measures are designed to keep the trusted system secure. The other is to develop a system that can function over an insecure network and implement a secure method for transmitting information. Because the Internet inherently is insecure, a trusted system is very difficult to create and it is even more difficult to get it adopted. A system that established authentication without resorting to being a trusted system is therefore preferred.
Digital certificates are evolving as a method for accomplishing both parts of the identity scheme. Digital certificates are the electronic equivalents of a driver’s license and a notary seal. Digital certificates are based upon public key technology, and X.509 has established standards for exchanging digital certificates. This allows different certificate authorities (CAs) to all issue compatible certificates, making adoption much easier.
Public key technology allows for authentication, provided that the user has the public key of the sender. Digital certificates are pieces of data signed by a CA that puts its reputation behind the data’s authenticity. The CA’s digital signature guarantees that no one tampered with the information, and their public key should be well distributed. As new CAs come into existence, their digital signatures can be added to a digital certificate issued by an already established CA. This allows a hierarchy of CAs to develop as well as a plethora of CAs, guaranteeing competition in this market.
Let us consider a typical interaction using a digital certificate. Suppose Bob sends Alice a digitally signed message. To authenticate Bob’s transmission, Alice must have Bob’s public key. Bob needs a method to send Alice his public key that would guarantee to Alice that it is his public key. Certificate authorities can solve this problem. Bob can have his certificate authority sign his public key. When a public key is combined with a trait and then the combination is signed by a CA, the item becomes a "certificate." Because his CA is well known, Alice’s computer knows the CA’s public key and can verify the CA’s signature. When Alice’s computer receives the digital certificate, she knows the information within it is accurate, and receives Bob’s public key. Alice can now verify that Bob sent her the message, and she can now send encrypted information to Bob.
Instead of creating giant repositories of public keys, trusted certificate authorities distribute their public keys, and sign the public key of users. Using the well-publicized Certificate Authority’s public key, Alice can verify that Bob’s public key is his, because Alice trusts Bob’s CA. After using the CA’s public key to verify Bob’s public key, Alice uses Bob’s public key to verify the message.
Because of public key technology, users can use the recipient's publicly accessible public key to encrypt messages. This way, in addition to confirming that the message was sent by the correct party and that no one has tampered with the message’s content, a user can guarantee that nobody but the intended recipient received the message. This is very important for e-commerce and sensitive private documents.
Although current certificate technology provides these benefits, the technology does suffer from some serious limitations. Use of digital certificates is not widespread, which limits its effectiveness. With certain products, like computer technology, the utility of a good increases as more people utilize the technology. And because the demand for the technology is likely to increase as more people adopt it, it is important to establish the technology with the early adopters.
The use of digital certificates is limited until large portions of users use it. Because most people don't worry about e-mail being insecure, there is limited benefit to digitally signing your e-mail. Additionally, encryption requires both parties support the technology, to know each other's public key. If the parties that you converse with do not adopt this technology, then you cannot send encrypted messages. Additionally, authentication requires that the other party is using software that can verify the authentication, and that the party is interested in verifying your identity. If your message's target isn't interested, there is no advantage to providing authentication capability.
Individuals use bundled certificates, like VeriSign Digital IDs, to sign e-mail and authenticate themselves to web sites that support VeriSign certificates. These certificates store all information about the user in one certificate, with different levels of certificates based upon the amount of information volunteered and the amount of real world verification that was done on that information. These certificates provide a one-to-one link between the person's real world identity and their online identity, as well as providing all their personal information to any server that they provide that certificate to. Companies also use certificates to sign their downloaded software packages and so that their servers can provide an encrypted connection.
The digital certificates of the status quo do not facilitate unbundling. Today, digital certificates are issued with all information in one certificate, and there is a dependence on the user to keep their certificates from being stolen or used by others. Additionally, most services only accept certificates from one certificate authority. Both of these limitations prevent current digital certificates from being used as an extensive digital identification scheme. Digital certificates can reap the benefits of unbundling, however, if they are designed with this capability in mind.
Every time a person purchases something with an age restriction, alcohol, tobacco, or pornography, they are asked to present proof of age. This proof can be visual verification; the individual is clearly not under 18 or 21. Alternatively, the age can be verified through a driver's license or other recognized form of identification. When someone provides a driver's license, however, to prove his age, he is providing more information than is necessary for the exchange. In the real world, we accept this because the person verifying your age is viewing the necessary information only for a few seconds, and carrying around multiple forms of ID to verify individual characteristics would be bulky and inconvenient.
Although bundling these traits in the real world often occurs as a matter of convenience, such bundling carries serious privacy concerns in cyberspace. In the digital world, all information that is made available can be and often is recorded. Additionally, the inconvenience of holding multiple forms of identity is trivial, with each digital certificate being a small file stored in the computer. If desired, picking between multiple forms of ID can be made trivial by automating most of the decision process in the client software. To accomplish this, a standardized language must be developed for identifying characteristics. A method must be developed so that two computers can communicate and request information, such as first name, age, birth date, age 18+, age 21+, citizenship, residency, etc., in a manner that allows the responding machine to locate and provide the appropriate certificate. This probably will require a technological standards board like the W3C to add new fields to the standardized language with each revision as they become necessary over time as certificate usage expands.
Certificate Authorities must begin providing multiple certificates when a form of digital identification is needed. Unlike a Digital Driver's license, it is important that when a user proves that he is 18, he does not provide the server with their name, address, social security number, e-mail address, telephone number, and birth date. Like the public key in traditional digital certificates, the CA’s digital signature authenticates that no one has tampered with the data. As certificates become commonplace, groups can begin issuing digital identities like they issue real space identities. The DMV can provide a digital certificate form of the driver’s license. To facilitate unbundling, the DMV could provide multiple certificates that include information like you name, address, birthday, age (including numerical, plus 18+ and 21+ flags), and other relevant information. This information can be presented as proof in cyberspace like your driver’s license in real space.
Currently, each server specifies which certificate authorities’ certificates it will accept. Because a digital certificate is a piece of data signed by an issuer’s certificate authority, we can create hierarchies of certificate authorities. Certificate authorities could sign each other’s certificates, creating a hierarchy that would allow a server to trust any certificate so long as some trusted CA in the chain has signed it. This should result in a system where a few large, trusted certificate authorities could serve mostly to accredit other certificate authorities, which would deal with end users. This approach would allow end users to face a competitive landscape, while still guaranteeing the credibility of certificate authorities.
To protect digital certificates, browsers or software applications could support a password feature. This allows users to keep other users of their computer from stealing their identity. This password protects the private key, and must be kept secret. If the password is compromised, the private key is compromised, and the digital identity is compromised.
The problem with this security regime is that it depends upon the end user to secure his certificates. There exists a risk that someone might steal the private key or that the holder of the key might disclose it to someone. There appears to be two types of technologies, if adopted, that would help minimize these risks. One is the implementation of a trusted system, and the other is an external verification system.
A trusted system that prevents the theft of the private keys could be built. A trusted system, however, has real world limitations. Requiring end-users’ machines to exist within this trusted paradigm seems problematic, as a sophisticated computer criminal could reverse engineer the certificate code, rendering the trusted system moot. The difficulties for building a trusted system on client machines is an extraordinary task, and convincing the pubic that the system is actually secure is an even more difficult feat. Further, a trusted system would not prevent a user from intentionally giving another access to his private key. Although trusted systems do have a role in digital encryption, they have a limited role in client-side adaptation.
The other system is to secure the private key with an external verification scheme. Instead of using a password, something not easily duplicated can be used. Smart Cards can store private keys and prevent tampering. Although a Smart Card is the size of a credit card, the technology is very different. Smart Cards store information in a chip, instead of in memory, and their interfaces are designed with specific methods of data transfer in mind. The Smart Card could also be used as a token to serve as the private key. Because it could not be easily duplicated, this would prevent users from giving their private keys to other users. The Smart Cards could be designed to not allow duplication through the card’s operating system. Extracting the chip to duplicate it will destroy the original, and the copy will fail. Smart Cards provide a reasonable level of security, reducing the likelihood of the public key’s theft or intentional misuse.
Another option is biometrics. Expect in the most extreme cases, this option would eliminate the possibility of public key theft or intentional misuse. Using a fingerprint, retina scan, or another physical characteristic of the real world to secure the private key and serve to authenticate that the user is the expected user. The person’s scan can be hashed to provide a value of the desired bit-length, allowing its usage of a private key to be used across borders. Alternatively, the scan could be used to provide a security shield for the private key, similar to a password but more secure.
However, any sort of identity architecture will require that people have access to a single system which they trust to be secure and not reveal their private key. To some extent this reduces the amount of security of any architecture and increases the burden on the user. Similar to the expectation that an automated teller machine will not store our card and personal identification numbers, an external verification scheme must not store data it reads for private key access.
Several mechanisms can be created to attempt to guarantee that a system operates in this secure manner. Manufacturers could sign the transmissions that are confirmed via scan, adding a level of credibility to their transmissions in addition to possessing the private key and certificate. A governing board that verifies that accuracy of these devices could also sign the manufacturer’s signature, so they could be universally accepted. If this system is standardized, and the method for verifying signatures from manufacturers, this system can be used without a trusted system. This requires, however, that each device be issued (or be able to generate) a private key and private key, a digital certificate, and the technology to digitally sign a device. While this does not require significant amounts of processing power from a computer, it might currently be beyond the capabilities of an inexpensive fingerprint scanner. This requirement could be added when these more advanced scanners are common and this would not be an unreasonable level of verification.
This allows us to decide the level of verification necessary within our system, while still maintaining a standardized system. We may decide that fingerprint scanners signed by manufacturers is too burdensome for selling pornography or tobacco products, but is reasonable for on-line voting. The server can set its requirements. This system can be adopted in any environment that requires authentication by applying additional levels of verification. The digital certificate system, thus, is extremely flexible in terms of additional security that can be used to protect private keys within the initial system.
The digital certificate regime described above facilitates unbundling, which is essential to maintain privacy in cyberspace. Cyberspace currently allows relatively anonymous usage, but once identity must be confirmed, all privacy is removed. The current scheme for confirmation is a credit card. A credit card releases information such as your name and billing address. When having Amazon.com ship a book to your home, this is adequate. However, when the desire is to prove your name to somebody in a chat room, or prove your age to purchase alcohol products, this system clearly releases more information than necessary. The digital certificate regime described in this paper can also be used to create unbundled certificates that link to pseudonyms.
Anonymous, secure, authenticated e-mail is another benefit of a certificate-based scheme. Users can use anonymous e-mail servers or re-mailers to create accounts with no link to their real world identity. Digital certificates, authenticating the user’s e-mail address and public key, like the VeriSign Digital IDs can work with an anonymous re-mailer. This allows user anonymity, protecting their identity, while prevent e-mail forgeries. Bob can create an anonymous e-mail address, and correspond with Alice. Although Alice will not know who Bob is, she knows that the same person is sending each of the e-mail messages, and she also knows that no one tampered with the messages or forged them.
Your digital certificates can include as many characteristics as desired, and they can be maintained separately. By unbundling them, information can be exchanged without revealing your real world identity. In this manner, users can remain anonymous, but authenticated.
It is very difficult to develop a guaranteed traceability feature for digital identity architectures. Systems that would store your real world identity in a certificate encrypted with a law enforcement public key have in the past been considered politically unviable. Each of the players in a transaction — the CA, the user and the web site store — could collude and avoid technologies or procedures that enable the government trace unless a trusted system with the trace built in is used. However, a viable option exists that could meet the government’s needs in a large percentage of transactions.
Essentially, the law enforcement trace is the product of the transaction. A customer must reveal his identity to the certificate authority for the CA to be able to certify any information about him. If the customer wants to purchase anonymously a product from a web site that requires him to prove his age, he would request his CA to issue him a new certificate certifying that a pseudonym is of the proper age. As a result, the CA always knows to whom pseudonyms belong. Similarly, the web site store always knows which pseudonym made a given purchase. To trace the identity, therefore, the law enforcement agent would have to contact the web site and demand to know which pseudonym is related to a known transaction, followed by a similar request to the certificate authority about the real identity associated with the pseudonym. This architecture would require the CAs and web site stores to keep logs that associate these items.
To implement this tracing, all Internet transmissions must at least require the use of a valid "null" certificate. Even in situations where a web site may not need or want the individual to present any certified information, a certificate still must be logged for traceability purposes. Essentially, a null certificate guarantees that the certificate authority is aware of the identity of the owner of the pseudonym, and that the Internet host is keeping track of pseudonyms, even when no identifying information is strictly necessary to the transaction.
The architecture can facilitate this with changes to the method in which web servers and other Internet servers receive their data, which will inconvenience users, but the system can be worked around to minimize the amount of differences the end user notices. The difficulty lies with coercing all sites to replace their server software. This initiative would require a rewrite of every Internet server, as well as complicating the server’s interaction with the operating system.
Most web servers store their data in HTML files on their hard drives. The directory-like structure of web sites allows users and webmasters to change HTML files with ease. The HTTP server that facilitates access to the HTML files (or any other protocol server providing access to data) can be easily modified to request additional identification before completing transactions requiring logging information. Only the web server need be modified in order to guarantee that this transaction information is stored; legal burdens placed on the provider of the web site would act as additional guarantees that such information would be stored and would motivate providers to make their servers compliant.
One of the advantages that a digital certificate based approach has is the decentralized nature of certificates. With a trusted system, one system must be established, and it is proprietary in nature. Digital certificates, like public key encryption, use a private key that is kept secret, and the public key and the certificate that can be distributed publicly.
Certificate based security also supports the concept of a certificate hierarchy. If Bob wants to send Alice his public key, he sends his certificate. If his certificate was issued by his company, Alice may not trust the certificate, because she only trusts a few services. Bob’s company may have negotiated a deal with VeriSign to issue certificates, and VeriSign could digitally sign all certificates issued by Bob’s company. Alice will now accept Bob’s certificate because VeriSign lent the certificate its credibility, even though Alice wouldn’t have trusted the company signature without VeriSign’s signature. Alice now accepts the company’s public key, and uses it to confirm Bob’s identity.
With hierarchies, a few large CAs will have their public keys widely distributed with all certificate aware clients. Additional CAs can issue certificates on behalf of these larger CAs on a contract basis.
Because digital certificates include more than just public keys, different CAs will serve different roles. VeriSign’s level 1 certificate includes a name or alias, e-mail address, and public key. This certificate can be used for authenticating that the e-mail came from the same person, but it fails to prove the sender’s identity. Their level 2 certificate includes their real name and mailing address. Digital certificate technology clearly allows a more flexible usage of digital certificates.
While the original usage of a digital certificate was to transmit public keys, there is no reason that other information couldn’t be transmitted, as explained above. The certificate based paradigm works on the assumption that large numbers of certificate authorities will exist to verify specific characteristics. This certificate signed by the issuer, gives credibility to the information contained within the certificate. These certificates can be transmitted over a secure connection to prevent other parties from intercepting the certificate and also verifying those facts against the transmitter.
Government CAs, or at least government contracted CAs, will play a large role in a working digital certificate regime. Information like age flags to confirm that the user is of legal age requires a legally binding certificate. This requires a licensing system for the issuing of legally binding certificates. Under X.509, the technical capability for transmitting interoperable digital certificates is available, so the process is now underway. A standardized language for the exchange of characteristics is the next step. Without that language, unbundling is made more difficult. The final technological hurdles will be reached once there is a strong consumer demand for these certificates. Without legal incentive to confirm identities, there will be no demand for the technology, and the necessary standards for these characteristic based identities will not be created.
By utilizing existing technology and the principles of unbundling, privacy, and accountability, we see the opportunity for a real, working digital identity scheme. A collection of protocols can govern these certificate exchanges, successfully protecting the private identities of citizens, while allowing other entities to confirm necessary traits on-line. A possible architecture would involve many servers communicating and protecting the privacy of their users.
Combining these ideas conceptually, we have developed a model for a hypothetical architecture. In our scenario, Bob wishes to purchase a product from the web site that requires him to be 18. Previously, Bob has previously obtained a "Bob is 18+" certificate from the CA. Bob enters negotiation with the web site, which wishes to know that he is 18. Bob creates a pseudonym, in this case "ghost," to shield his identity. The pseudonym is then signed by his CA indicating its authenticity. The CA now knows that Bob is attached to the pseudonym "ghost". Bob can then request from the CA a "ghost is 18+" certificate. Bob then engages with the web site store, sending them the signed public key for ghost (alternatively, Bob can provide a public key server with ghost’s signed public key and the web site can obtain it from the public key server) and the signed "ghost is 18+" certificate. Using the signed public key, the web site can verify that the user presenting ghost’s public key is the holder of ghost’s private key. Successful verification indicates that the transaction can be completed. The key features of this architecture are that the web site can verify the needed information without knowing the real world identity of ghost, and the certificate authority knows the real world identity of ghost, but not the transactions in which ghost engages.
The business world is also very interested in the development of a digital identity authentication system. Electronic commerce, with its ability to market and sell products directly to the consumer at lower costs, promises to become a key channel for retail in the twenty-first century. The success of initiatives such as Amazon.com has caused a flurry of development of World Wide Web storefronts for traditional retailers of products ranging from books to computers. Taking a cue from the work of direct marketers, corporations have begun to recognize the Internet’s potential to facilitate the tailoring of the online storefront to each individual customer. Purchase a book from Amazon.com, for example, and on subsequent visits to the web site you will be presented with the titles of other books that you might like on subjects similar to the one that you purchased or other works by the same author. Such tailoring depends on the ability to connect bits of information to the identity of the visitor. This section examines how the proposed digital authentication mechanism might be put to use for commercial activities by businesses and consumers.
Previous sections of this paper examined whether the architecture of a digital authentication mechanism should be designed to permit traceability. Although the discussion focused on how traceability on the Internet would meet the needs of the government in carrying out its law enforcement function, it should be noted that businesses also have an interest in the development of an architecture with such a feature. Many corporations have established intranets to facilitate communication between the various divisions of their companies. Traceability in the architecture would help the leadership of a business monitor the activities of its employees. Monitoring of this sort might be motivated by a desire to track the productivity of individual workers or a need to ensure procedures designed to govern access to the company’s sensitive information are followed. The development of an architecture for the Internet that included traceability would provide a standard that could be adopted for corporate internal networks, without the associated research and development costs.
Aside from the caveat presented above, business-domain interests in the use of identity do not require developers of the architecture to make any fundamental architectural choices for the system. Instead, most of the concerns regarding the business arena are related to how businesses and consumers will behave in an environment using the digital authentication mechanism proposed in this paper. The remainder of this section shall consider these issues as they appear in the business-consumer and the business-business relationships.
Both consumers and businesses stand to benefit from the implementation of a digital authentication mechanism similar to the one proposed in this paper, despite the sometimes conflicting interests of these parties. This subsection shall examine how consumers and businesses might choose to use a digital authentication mechanism and what they might layer on top of this base architecture in order to better meet their interests. The discussion is separated into two parts, examining first those issues stemming from the ability to authenticate information and then those issues that proceed from the unbundling of identity information that the digital medium allows.
The ability to send authenticated information, and the other party’s ability to verify this authentication, is valuable to both consumers and businesses. Consumers might wish to demand that a corporate web site provide it with a digital certificate in order to prove that the web site truly belongs to the corporation. Although many consumers do not question the authenticity of the web sites they visit, there have been reported incidents of individuals using web sites to commit fraud. These web sites may be designed to mimic the web site of a legitimate corporation or may purport to represent the web storefront of a business that in fact does not exist. The practice of confirming that the web site is indeed the web site of the business it seems to represent will grow in importance as consumers purchase a larger percentage of their goods and services via the Internet. Under the architecture proposed in this paper, a business could make available on its web site a digital certificate, for example, signed by the Better Business Bureau Online or by the CEO of the corporation. The consumer would download this certificate and verify the signature before interacting further with the web site.
Although the architecture proposed in this paper would facilitate the construction and exchange of such certificates, it is an altogether different question whether businesses will agree to create them. Certificates designed to authenticate that the web site is the corporation’s legitimate Internet presence are likely to be adopted by companies with little opposition should the idea be presented to their executives or the need made apparent by consumer demands or publicized incidents of fraudulent commercial web sites. These certificates serve to protect the reputation of the company’s brand name, one of its most valuable intellectual property assets. One might see voluntary participation in programs certifying other corporate practices if the social and market pressure is viewed to be large enough to warrant the expense of finding a trusted third-party to verify that the corporation in fact complies with its stated practice. For example, Ben & Jerry’s advertises that it does not use milk containing Bovine Growth Hormone in the manufacture of its ice cream. It may be cost prohibitive, however, for Ben & Jerry’s to pay a third party to vouch for this fact. The third party may agree to authenticate this statement only if Ben & Jerry’s permits auditors to test the milk on a periodic basis. The costs of such an arrangement would be passed along to Ben & Jerry’s as part of the cost of the authentication. Where certifying certain business practices seems crucial and has become an issue of regional or national importance, legislatures might pass statutes mandating the use of digital certificates. Consumer data privacy, mentioned above, is one area that might receive such consideration in the future.
From the corporation’s perspective, the ability to receive authenticated information about its customers is very useful. Many industries operate under state and federal regulations which dictate to whom they are permitted to sell their products. Vendors of alcoholic beverages, for example, are prohibited from selling their products to individuals under the age of twenty-one. The inability to verify this information inhibits the development of electronic storefronts for these industries. Similarly, under pressure from state governments, companies might demand that consumers present digital certificates certifying their place of residence so that the purchases could be taxed at the applicable jurisdiction’s rate. Currently, businesses often request that consumers provide information such as their occupation or income level when making a purchase. Typically, no purpose of the transaction is facilitated by this information other than the company’s interest in knowing as much as possible about its customers. Businesses might begin requesting authenticated versions of consumer information as this would make their profile databases a more valuable commodity than those without the degree of authenticity a verified digital certificate provides.
Whether consumers will comply with the demands of business and produce these certificates largely depends on the balance of power between the company and the individual in the consumer negotiation. Where the government proscribes the sale of a product to a certain class of individuals, consumers wanting to purchase the product will have to present the appropriate digital certificate as the business cannot sell the items without identifying that the purchaser is eligible under the law. Concerning other demands for authenticated information, factors such as the individual’s personality, whether the average consumer provides the information, and the importance or necessity of the product to the consumer. Unless the government steps in, it is impossible to predict whether the business will lower its demands or the consumer will have to forgo purchasing the product should he not wish to deliver to the company the demanded information. The next section explores this issue in more detail, although from a slightly different angle.
One of the advantages that the digital medium provides is the ability to unbundle portions of your identity that otherwise would be connected. In real space, a customer purchasing a book in a store reveals more than just his desire to own that particular novel and his ability to pay the store’s price. His physical presence in the store reveals his gender, his height, and his taste in clothing. Being a digital medium, cyberspace promises to allow for each of these aspects of identity to be represented separately and displayed to others only when needed. This subsection examines whether market pressures might prevent users from taking advantage of this benefit of unbundling.
It should be noted that underlying this entire discussion is an assumption that consumers care about privacy. As presented earlier in this paper, the degree to which an architecture facilitates unbundling determines the amount of privacy that the system permits. A person’s privacy is best protected by never requiring them to reveal information unnecessarily. Procedures that require those who collect information from individuals to safeguard it are only second best. The digital authentication mechanism proposed in this paper allows a consumer to separate the various elements of his identity, giving him the capability to reveal only what is necessary for each business transaction into which he enters. It is not easy to predict the extent to which consumers will take advantage of this feature of the system that protects the privacy of their identity. In part, this stems from the fact that most individuals have no past experience in which they had a similar level of control over what constitutes their identity. Which pieces of information are included in a driver’s license, for example, is set by the government. A certain amount of acculturation to the architecture inevitably will be necessary. Further, the level of privacy each person requires varies. Erring on the side of perfect unbundling on the architectural level seems appropriate given these circumstances.
The level of unbundling that occurs within such an architecture, however, depends on whether people choose to unbundle their identities. People are subject to the social pressures of their peers and the demands of the marketplace. These pressures may lead individuals to bundle portions of their identities that they otherwise would unbundle absent these outside influences. Theoretically, the pressures of the market could be so great that there might be more bundling of identity information under this system than exists today, despite the potential of the architecture.
Figure one, shown above, helps clarify the possibilities of the architecture in light of the pressures of the marketplace. There exists today a certain level of unbundling, represented in the figure by a bar on the spectrum. The digital authentication mechanism’s architecture facilitates the possibility for the bar to be shifted to the right as far as each individual desires. However, businesses might demand that certain bits of information remain bundled and be revealed in order to complete a business transaction. Such market pressures push the bar back to the left towards perfect bundling. Where should the bar end up resting on this spectrum? Although, it’s possible to make various arguments for different positions within the area of the spectrum labeled as B, it might be best left to the marketplace to work out the exact position. One can argue, however, that market pressures should not be allowed to push the bar into the section labeled as A. The points of the spectrum labeled as A were options available, but not chosen, under the status quo’s architecture. As bundling is an option under the status quo, one can surmise that consumers prefer at a minimum the level of unbundling provided by the status quo. Should the bar be pushed into the section of the spectrum labeled as A, policymakers should consider legislative remedies to reposition the bar in favor of unbundling.
There is reason to be concerned that in the future businesses might pressure consumers to rebundle identity information so that it is bundled to a greater degree than today. There exists a valuable market for consumer profiles. Businesses desire two types of information from consumers: direct feedback on the company’s products and information identifying the consumers’ preferences or traits. As it is difficult to survey or measure consumer preferences, business substitute consumer profiles, or identity information, correlated with purchases to develop a picture of what motivates an individual to buy a product or which types of products an individual is likely to purchase. A company concerned with avoiding unnecessary expenses would prefer to purchase or lease a consumer profile database if such a move would cost less than it would to compile its own database. As it costs next to nothing for the business creating the profile to sell or rent a copy to another corporation, there exists a healthy market for consumer profiles. The more complex and detailed the profile, the more valuable it is. So long as this market exists, all other things being equal, companies shall be motivated to demand as much information from the consumer as possible.
Businesses are likely to design their request for identity information so that it appeals to the financial considerations of the consumer. Were a business to refuse to sell products unless the consumer provide identity information to the company, the business would run the risk in the macro view of selling no products, or an insufficient amount to stay in business. Thus, the selling of its consumer profiles is inherently a secondary source of revenue for the company. Instead, companies might offer discounts to customers who provide a digital certificate certifying various bits of identity information. This occurs regularly today. Among other things, an application for a store discount card often asks the customer to provide his birth date, social security number, and household income level.
Some are sure to argue that this commoditization of identity information is appropriate and desirable. By providing a price for privacy, consumers will evaluate their preferences for the "good" and purchase the desired amount of privacy. Besides arguments that criticize this purely economic view of identity, there exist reasons to doubt whether a consumer fully considers the ramifications of providing a company with information about his identity. Does he take into account when calculating his preferred choice the potential third parties that could purchase the information and use it in a way that affects him negatively? Does he take into account that the bits of information that he might choose to reveal to company A might be combined without his knowledge with those bits of information that he revealed to company B? A consumer who is comfortable with company A knowing ten aspects of his identity, but no more, must understand that the only way to ensure company A never acquires more than those ten traits is to never reveal to any other corporation a different set of traits.
Ultimately, it is up to the judgment of legislators to determine when the government should step in and regulate the type of information a business is permitted to collect. All that can be said for now is that legislators probably will take the approach of proscribing the collection of particular bits of information rather than the more intrusive and difficult measure of identifying the categories of identity information which may be collected by commercial enterprises.
The Internet is far more than just a place where businesses shall sell their products to consumers. As a communications technology, the Internet also facilitates communication between businesses. Today, most corporations contract with other companies to supply them with necessary components or raw materials for their products. For example, the major automobile manufactures outsource the construction of many of the parts necessary for their vehicles. In such situations, it is important for the corporation to be able to communicate with its supplier the specifications for the component. Often, this may require the company to permit the supplier to access documents containing corporate trade secrets. Two companies coordinating research and development efforts would need to exchange such information frequently. The Internet provides a low-cost method of delivering this information to the supplier. Because of the sensitive nature of the information, however, the ability to authenticate the identity of the individual attempting to access the documents as the approved supplier is critical.
It should be noted that the ability to authenticate information is the aspect of the proposed architecture that aids the development of business-business relationships on the Internet. Whether the digital certificate certifies that the supplier is "authorized to access file X" or that the supplier is "company Y" makes no practical difference. Both parties to the agreement already know the other’s identity. It is difficult to think of what interest a company might have in not revealing its full identity as its reputation and brand name is one of its most valuable assets.
Cyberspace has an important social aspect to it that must not be overlooked. Ever since the ARPA Net was created its primary use has been to communicate with other people. With the advent of a faster backbone, different types of communication media became possible — namely, interactive communications. Community in cyberspace is based on the interaction between people.
Although a community is a group of people who interact with each other, at the basic level it comprises a group of people who exist with each other in a common plane.Cyberspace can be treated as a conduit touching portions of real space at key points. Ideas are passed through the conduit, and business is transacted through this conduit. The cyberspace community are members of the global community interacting on a different plane than in real space. These members rarely interact in real space, but they communicate through multimedia means in cyberspace — whether it be by text, image, sound, or a combination of the three. It is not possible to use the Internet without being part of this community of people; you cannot avoid being a part of the community, even if you are using the Internet as a conduit: by e-mailing people, reading web pages, reading newsgroups, or doing commerce online, one has joined the cyberspace community.
For example, communication evolved into allowing people to meet up in multi-user dungeons (MUDs) to interact with other characters that people create to represent themselves. This concept of the MUD is the basis of the community in cyberspace. In the community created in a MUD, every member puts on a "mask" to pretend they are of a certain type of person (depending on the setting of the MUD, it can be anything from a serial killer to an ogre), and then they run free in this virtual world. However, the everyday usage community on cyberspace has not fully evolved to this state yet. One does not yet see avatars of people wandering, meeting, and interacting with other avatars through cyberspace. On the other hand, cyberspace is more than a collection of people who merely exist together.
The Internet newsgroup is a prime example of community. A newsgroup is a place where people with the same interests can post messages and allow people to read and respond to these certain messages.
Examples of these are never ending: there are those newsgroups which specialize in maintaining certain operating systems on computers, those talking about northeastern birds, and those which contain erotic fiction. People cluster and interact on these newsgroups; they are free to debate politics or philosophy on these newsgroups while others either listen or cheer them on. Currently, anybody can post on a newsgroup and can post either with his real name or an assumed one.
Most post under their true e-mail address, but when put under consideration, there is no reason to post a message with a name connected to it unless he wants credit for a post, or he expects a response. Nobody uses this "anonymity" feature, however, unless they are posting something questionable and do not wish other people to track them.
The creation of a web page is also a symbol of community. The web page is an invitation to interaction. Personal web pages usually contain the hyperlink saying "e-mail me!", thereby spurring on communication. It is similar to a sign on the door of a house saying "Welcome to my home." A web page is that front door sign — it is inviting the person in and introducing the author to the visitor.
Doing commerce is another sign of community. In this community there are those who create the goods that others want to buy. The Internet simply provides that conduit between the buyer and the seller — all is needed is for the peddler to place his web page up inviting the buyer in for the interaction, and then the good can be sold online.
It is not usually possible (not due to technicalities, but due to situation) to do commerce anonymously online. The only good available online currently that can be purchased anonymously is software that one can download after purchasing. Any other commerce requires a real space address for the shipment of the goods.
Cyberspace provides all these means (social and commerce) for people to get together and interact on all these different levels with each other, not only to exist with each other. And in this current cyberspace community, there is the option to work anonymously, or there is the option to connect some identity to yourself.
Anonymous transactions on the Internet today are not always possible, and when they are, they are not trivial to perform. There are many anonymous re-mailers around the Internet — anonymous re-mailers provide the way for people to send mail or to post to a newsgroup without having their identity (their e-mail address) to be attached to that post. This provides the service of sending an e-mail or a post "one-way." It is very similar to the idea of the anonymous pamphlet distribution. One can leave an information pamphlet at your door that simply contains information but no point of contact back to the distributor. In order to facilitate some two-way communication, it is a little bit more difficult. It is possible to create a server that will re-mail your e-mail anonymously, but provide a handle on the e-mail so when another attempts to reply to that e-mail, instead of disappearing, the server actually resends the reply back to the originator of the first e- mail. This of course provides the re-mailing server with all the information on who is who in this system.
However, there is no way to do anonymous commerce on the web currently — all commerce requires you to provide some form of payment which is either in the form of electronic money (credit cards), or a real world equivalent which needs to be transferred between the two parties.
On the other end of the spectrum is the absolute identification of oneself. In order to facilitate this, one needs to use some form of digital authentication mechanism such as a VeriSign ID, or a PGP key — this ID can be verified to be the real thing either by VeriSign, or from a PGP key server. For an equivalent of certified mail, one would like to know whether or not the sender is actually who he claimed to be. With an e-mail or a news post signed with either the VeriSign ID or the PGP key, one can be absolutely sure that the sender is who he or she said he or she was.
Both of these are optional when interacting on the Internet today. However, if full unlinking of identity or absolute linking is enabled , in cyberspace interactions, then there are ramifications in the social arena that must be considered.
Full unlinking provides what many consider to be the ideal intellectual society — freedom of ideas is possible. Unlinking allows thoughts to have a "life" of their own; thoughts gain separation from the person. Ideas and concepts flow from person to person without necessarily tying themselves to any one person.
Unlike in cyberspace, in real space it is impossible to facilitate full unlinking. In real space, it is impossible to interact with other people and not leave a portion of one's identity behind. In contrast, the consequence of digital communication in cyberspace is that nothing remains connected to the original source.
In a cyberspace with no mandatory traceability, no one need fear that posts intended to be anonymous could be traced back to them. This allows a full range of speech, including anti-government speech, without fear of repercussion from an oppressive government. As long as one can express what he wishes to express, one does not have to worry that he exists in a Orwellian state. For example, one could post an anonymous message to a bulletin board stating that the U.S. President should be removed from office, and not have to worry that somebody may be able to track his name back from that posting and be able to find him and cause him personal damage. One can say whatever he wishes to whomever he wishes whenever he wishes.
Unlinking also protects the "whistle-blower" who leaks useful information that may get him in trouble had he posted it directly with a connection back to his identity. There may be many instances where information has not been released because the potential whistle-blower feared reprisal. In the absence of this fear, an individual will be more likely to release this information to benefit the greater good.
Unfortunately, full unlinking also provides an environment for a large amount of "junk" posting on the Internet. It allows people to post false and fictitious ideas without fear that they will be told to stop since there is no way they will be able to be tracked down. Problems of this type have occurred on bulletin boards — people have posted false information concerning certain stocks, which caused readers to react in a way that affected the entire market. And due to the unlinked nature of the conversation, there is no way to trace the source of the false information down and prevent further incidents. It is posters like this who give Internet bulletin boards a reputation as untrustworthy places to obtain information. The consequences extend far beyond simply damaging the reputation of the Internet; postings such as this will affect the real world.
The "junk" being posted is due to the elimination of social norms. If nobody is able to trace one's actions back to their real world identity, then there is no incentive for one to control his or her activity; no longer are norms regulating what people do in cyberspace. One has the opportunity to do whatever he wants and not have anybody "slap his wrist" to restrain him.
This ability to fully unlink also will make the Internet a prime candidate for crime. It is possible for sections of cyberspace to appear, whether it be an Internet newsgroup or a web site ring, for the specific purpose of trading criminal information.
Criminals will have no fear of being caught and prosecuted for these postings since there is no way for the message to be tied back to the real identity of the server. Nobody will necessary know any aliases being used since nothing has an absolute link back to the real world identity of a person. This alias can simply be set up for use in this bulletin board system. Although other posters may know to whom this alias refers, to the uninitiated the alias will mean nothing. Therefore, the law enforcement force will have no way to track down those who may have posted these messages. Nothing stops criminals from coordinating over the Internet, or even posting criminal messages on public bulletin boards.
Law enforcement has lost a key tool for tracking down criminals. Simply having a criminal newsgroup post is not enough; there is absolutely no way to trace the real world origin of this message. The only resort is to attempt to analyze the content of the message. This can be easily circumvented with careful measures, which may not be that intricate. As a result, a criminal can communicate with other criminals and have no fear that anybody may be able to track him down simply for posting to a newsgroup. Already, strong encryption sometimes prevents law enforcement from determining the contents of a message. Unlinking exacerbates the situation by making it impossible for law enforcement to determine even the sender or recipient of the message. In the real world, law enforcement can use telephone records to investigate criminal activity; however, mandatory unlinking works completely against this. This means current law enforcement measures become less effective.
Should an Internet, which easily facilitates, if not promotes, crime be allowed to exist? The crucial question becomes which choice is more important — anonymous access for everybody everywhere with the side effect of promotion of crime, or a mandatory real-world link associated with information for the sake of safety.
The other side of the spectrum must also be fully considered — the mandatory linking of identity with each cyberspace transaction. In other words, each e-mail sent, each newsgroup post, and each commerce transaction will have a digital feature that has the ability to be connected back to the transactor. The government can identify the originator of any newsgroup post — not simply the alias which posted it, but the real world of the creator.
This again is not a true analog to the real space domain. The few mappings include phone calls and credit card transactions. Each phone call is logged for billing purposes and can be obtained by other third parties if they have the proper credentials in order to do so.
In cyberspace, e-mail and all digital transactions can be hidden and secured with encryption, leaving third parties no more information except who the call was made by, the time, and to whom. All e-mail and newsgroup posts are marked exactly with the time they were sent, and with the mandatory linking of digital transactions, the information of the parties involved in the transaction will then be included; however, the content of the message can be hidden.
This situation will eliminate the ability of anonymous re-mailers to provide their service fully. It is not to be said that anonymous re-mailers will not still be used — all that can be hidden is the given e-mail address of the originating party. This may be enough for the digitally unoriented, but law enforcement could still access the identity link of the transaction, and the true identity of the sender could still be determined. If anonymous re-mailers are allowed to exist in this situation, then they must be designed in order to account for the "filtering" effect of the e-mail. E- mail may be able to be passed on from entity to entity but the identity of the originator must be included somewhere in the transaction.
Whether or not this will cause a major change in the philosophy of usage in cyberspace is an interesting question. The item of interest is who holds the ability to check the identity of the transactions in cyberspace. In this model, only the government has the ability to absolutely identify transactions. This will probably mean nothing to most transactions on the Internet — standard social transactions or commerce will be unaffected unless they are potentially criminal transactions.
This traceability means nothing to the end users of cyberspace, except if they are engaging in illegal activities. The law enforcement agencies will have the ability to absolutely identify people who are communicating with each other. This eliminates the fear that cyberspace may not be patrolable. Law enforcement has gained the ability that they have in real space with phone records. The officers still do not have the ability to necessarily tap into the content of the transaction or the interaction, but they do have the ability to see who is talking to who.
Now, law enforcement's reach has stretched out into commerce. When I purchase something online, the record of the transaction is available for a third party to view it. This will cause social repercussions. This implementation is a mandatory enforcement of what social norms should be taking care of. No longer are people relying on their own guilt to govern their lives; no longer will people have the security when they buy something that "nobody else" will know that they have spent money on a certain good or service (of course this includes not only illegal items, but legal yet socially unaccepted goods). Somebody somewhere may have access to the transaction record and may view it.
This may prove to be damaging to cyberspace commerce. If a precedent is set to create a mandatory link between identity and cyberspace transactions, then other commerce centers may decide not to have this mandatory link, causing some consumers to use services in different commerce communities outside the United States. This will begin a downturn in cyberspace activities in the United States.
Unfortunately this also provides an incentive for criminal networks to start their own private internets for their own use. These networks will have their own security, their own group of people who are allowed to use them — and, most importantly, no identity linking. This allows criminals to evade law enforcement and all the effort being placed into setting up the mandatory link, making efforts to establish identity linking seem like a waste of time. However, if this digital identity system is not set up, then cyberspace will be used for crime
As one easily can see, this evolving situation seems to be leaning toward a "big brother" type of scenario. That is why the statement "standard social transactions, or commerce will mean nothing [to anybody] unless it is questionable" is so interesting. Who or what defines "is questionable"? This argument seems similar to the random drug testing argument — if one does not take drugs, why should one mind being randomly tested for drugs? The parallel to this argument is if someone is not sending something illegal over the Internet, why should he mind that his absolute identity is being tagged on every transaction in cyberspace? The same applies for commerce — if one does not purchase or trade anything illegal in cyberspace, why should one be concerned that somebody may have a record of all the transactions performed.
This type of reasoning is incorrect — one's right to privacy should not be jeopardized for the convenience of law enforcement. In general, people should be able to act without worrying about somebody looking over their shoulder at every step. A mandatory link might stifle social activity on the Internet as individuals may view the link as a substantial invasion of their privacy in cyberspace.
Granted, this puts a burden on the law enforcement community to attempt to keep order in a digitally oriented culture by not providing them with a precious tool: the ability to precisely identify a community member's real life identity in order to prosecute the source of any disorder in cyberspace. This may be the best that law enforcement can hope to do in cyberspace.
The current state of cyberspace identification mechanisms is far from the flexible, broad potential of the identity architecture we have proposed. There is still a long way to go from the ‘here’ of the Internet as it exists in 1998 to the ‘there’ of the ubiquitous, secure identity architecture we have described. In order for the Internet to reach its full potential, a secure mechanism for managing and verifying digital identity is necessary. There remain a range of hurdles to overcome before a cyberspace identity mechanism will be deployed and ubiquitous. These hurdles can best be analyzed in four categories: social norms, market, legal, and architectural barriers.
The main social obstacle to implementation of a cyberspace identification mechanism is that the general public does not recognize that there is a problem with the existing identification architecture. The general public does not understand the need for an improved, secure cyberspace identification system. Even without any effective identification mechanism, the use of the Internet — for both casual and secure applications — has soared, with double-digit growth rates measured month-to-month rather than year-to-year. While more sophisticated Internet users may recognize the need for a digital identity mechanism, these advanced users represent a shrinking percentage of the overall Internet ‘community.’ Many people using popular Internet applications seem to be satisfied with the existing levels of security and identification. E-mail, for instance, is often self-identifying through the content of the message. Forged e-mail, while easy to create in the current architecture, is not perceived to be a major problem. E-mail eavesdropping, also a relatively simple technical task, has not slowed the flood of e-mail communications. On-line commerce is booming even based on systems requiring credit card numbers and the overly revealing identification that credit card numbers enable.
More advanced uses of the Internet, however, are suffering from the lack of an identification architecture. There is a wide realm of future cyberspace applications that cannot be facilitated without secure identity verification. For example, "registered" e-mail that provides proof of sending, proof of receipt, and a chain of custody, requires a secure identification mechanism. Public Internet terminals that, regardless of their location, can configure to a user’s preferences and load a user’s personal data require secure user identification. Likewise, important applications such as on-line voting in government elections or filing on-line tax returns with the IRS will demand an effective authentication methodology. The general public, however, does not yet comprehend the need for an identification architecture to enable these advanced applications. Most cyberspace users are content — even amazed — by what they presently can do online.
The lack of public understanding about the identity problem is increased by the potential to unbundle digital identity. Identity in cyberspace need not be — and should not be — merely a direct translation of real world identity. However, changes such as unbundling identity into discrete traits and providing multiple identities to the same person require the public to comprehend their own identity in new and complex ways. This is a slow and difficult process.
The key solution to this social barrier is to educate the public about the need for a cyberspace identification architecture. Until people understand that existing cyberspace identity mechanisms are insufficient, people will not begin using more secure identity mechanisms and will not begin lobbying the government to pass legislation encouraging an Internet identity system. This burden of education falls on companies that are trying to develop identification mechanisms, such as VeriSign, Entrust Technologies, and Zero-Knowledge Systems. For instance, according to VeriSign CEO Stratton Sclavos, VeriSign is giving away most of its consumer identification certificates for free in an effort both to train consumers about the need for cyberspace authentication and to develop the VeriSign brand. However, much of the rest of the business community developing Internet services and software currently is promoting the message that "the Internet is safe," in an effort to encourage electronic commerce and the other presently available Internet services on which these companies rely for their revenues. Such conflicting messages delay consumers’ learning process about cyberspace identity and reveal the need for advocacy groups to promote the importance of a secure cyberspace identification system. The educational barrier may be the most significant barrier for consumer identification technology. According to Mr. Sclavos, "consumer markets are just not ready — the technology is there, but consumer behavior is not."
The market barriers to the implementation of a secure Internet identification system stem from the difficult business economics inherent in solving this type of problem. One of the key problems is that there is significant business model risk for companies providing identity verification solutions. In other words, it is unclear exactly how these companies can make money. In addition, economic incentives do not encourage the development of an open-standard identity infrastructure. Ultimately, success of an open-standard identification architecture, such as our proposed system, may require government intervention in the marketplace.
On the revenue side, revenues from identification services are likely to be very low. People are not used to paying for their own identity, beyond a nominal amount. For example, even a U.S. passport, perhaps the most secure and broadly accepted of all forms of identification, costs just $60 for a 10 year term. For $6 revenue per year per user, it would be difficult for a private company to become profitable. The only way that the U.S. government is able to offer passports for this low fee is that taxes and other government revenue subsidize the costs, and the government is not seeking a profit. In addition, unbundling, which is one of the key benefits we have described for an advanced cyberspace identification architecture, may potentially drive revenues from identity services even lower. It is possible that the more that specific traits of identity become unbundled from each other, the less a consumer is likely to pay for verification of each particular trait. However, this is still very new ground for both consumers and businesses, so it is also possible that unbundled identification traits could actually be more valuable and generate higher revenue than bundled traits, as consumers’ privacy remains protected and the disclosed trait could serve a focused and desirable function.
On the cost side, building the infrastructure for an identity solution is very expensive. The system has to be both highly secure and operational, with a fast response time, on a 24-hour per day, 365 day per year basis. This requires a significant amount of capital to be invested at the beginning, in order to set up the service. In effect, the entire system must be built — at least in a small-size version capable of being scaled up rapidly — before there are any clients using it. This pattern of cash flows make the identification system business particularly unattractive.
Even if a profitable business model can be developed, it is unclear which company, or even which type of company, is most appropriate for building an identity infrastructure. Which companies should issue certificates? Which company is trusted for this highly secure and important function? How many CAs should there be? CAs could be either new companies specifically started for this purpose, such as VeriSign, or could be trusted existing companies, such as AT&T or the U.S. Postal Service.
To understand who is best suited to develop the infrastructure, it is first essential to understand that the identity architecture would be, in economic terms, a public good. This is true because the architecture would be both non-excludable and non-rivalrous. First, the ideal identity authentication infrastructure would be an open standard that allows particular users to customize identity applications for use in a wide variety of contexts. As an open standard, the system would be non-excludable, or in other words there would be no way to exclude someone who does not pay for the open standard from enjoying it. This type of identity system would be an improvement over closed, proprietary systems, because the benefits of the architecture rise as the number of compatible users increases. In contrast, currently available identity mechanisms tend not to interoperate. For example, the various internet telephony applications that are available all allow voice communication over the network, but only among users of the same brand of telephony application. Similarly, an identity mechanism provided by one company might be incompatible with a mechanism provided by another company. Second, our proposed identity architecture is non-rivalrous, meaning that one person’s use of the system does not decrease another person’s enjoyment of it. The system, in effect, is not used up by any particular person’s use. In fact, the opposite is true: there is a rising marginal value of each user of the identity architecture. As the number of people using the same identity architecture increases, the value of that architecture increases, because each identity application based on this architecture would be compatible with other applications also based on the architecture.
The fact that an identity architecture is a public good both reveals some of the current difficulties with developing this type of system in an unregulated market and highlights what is necessary to implement this system successfully. First, as a public good, there is a disincentive for any particular private company to develop a truly open standard identity architecture, and there is a corresponding incentive for each company to build proprietary identity architectures. This is because the company that develops an open system does not necessarily get to capture all of the value that the system creates. Instead, a significant portion of the value accrues to other companies and other parties that develop applications for their own use based on the open standard that has been created. This pattern helps to explain why almost all of the presently available identity architectures are proprietary standards that are not interoperable. Each company is hoping both to develop the standard and also to capture all of the resulting value. No company is willing to create the excess societal value that emerges (but that does not accrue to the developer) from an open standard for identity verification that allows global interoperability. The TCP/IP standard provides a good example of this economic incentive pattern. With separate, incompatible network protocols, society suffers due to the costs of maintaining separate networks, the burdens of network interchanges, and the loss of shared information. When the TCP/IP standard emerged to help produce a single global network, society benefited from this public good. Enormous wealth was created, but no single company alone captured the value from deploying TCP/IP or charged for the use of this networking standard.
Due to this disincentive for any company to produce a system that will be open and interoperable, it appears necessary for the government to regulate or at least to influence this marketplace. An open standard identity infrastructure — a public good — would produce benefits for society beyond just for those who create it, and thus private parties are likely to invest below the socially optimal level. While the potential for industry to coordinate investment efforts and develop an identity system exists, such coordination might not take place. If the public wishes to ensure that an identity system is developed, then it the government is economically the best party — and probably the only party — which could guarantee the building of an open identity architecture. In addition, as we have argued above with respect to the government’s role as law enforcer, it may be necessary for the government to influence the identity architecture now — before the identity mechanisms are determined — rather than to wait until later, if the government’s legitimate needs for forced revelation of identity in certain limited contexts is to be enabled on the Internet. Finally, a secure identity verification mechanism could be extremely valuable to the government for purposes of national defense and national security. Whether the government is motivated by economics, law enforcement, national defense, or a combination of all three, there are several ways in which it could trigger the development of the identity infrastructure.
Initiating an identity infrastructure first requires the development of the architecture specification. This involves writing the certificate specification and protocol for exchange, and ensuring that both are robust. Next is the implementation of software which uses the specification, and we conclude with the deployment of these applications to the consumer market. It is likely that any company which invests the time and money to implement applications for a specific architecture will make these products available to the consumer market.
The U.S. government has four basic options if it wants to initiate this procedure. First, the government could decide to build the system itself. A new or existing government agency could have responsibility for developing the specification, and the government, via legislation or market power, could then influence Internet companies to develop identity applications using this specification. While this scenario might seem simple, it is unlikely. The government is not particularly well suited for this type of technology development. In addition, this is much more of a ‘command and control’ method of governance than the U.S. has normally adopted. Particularly with respect to the emergence of Internet technologies, the U.S. government has so far taken a free-market approach to technology standards or at least deferred to industry associations. Indeed, given today’s political environment, the first option may be both undesirable and unrealistic.
Second, the government can contract with a private sector organization to perform the specification development. This makes more sense than internal government development because the system ultimately has to be implemented and maintained by the private sector and must be compatible with private sector products. Even if the government funded the identity architecture development, the issue of implementation would remain. Implementation could be mandated by legislation or promoted through financial rewards or liability avoidance.
Third, the government could encourage a single company to develop the specification for the identity architecture by granting to that company a monopoly right such as the right to license the open standard specification to software developers. Note that this would have to be limited to a form of monopoly that did not discourage people from adopting the open-standard. For example, the company might charge a nominal licensing fee for developers; too high a fee would result in industry backlash and the ultimate demise of the standard. Additionally, the company might not be the best in the market. If the government wants to guarantee adoption of the standard, they will additionally need to regulate the implementation stage, as in the second approach.
Fourth, the government could pass laws assigning tort liability for identity misuse, while avoiding the role of technology developer or technology financier. In this context, the government is allocating liability in order to encourage entry into the market; allocating liability as a method of encouraging use of the system will be discussed below. This approach fails to address the problem of interoperability: companies likely will select the cheapest system, even at the expense of interoperability. Unless the private sector has a specific incentive to develop interoperable systems ¾ an incentive which liability rules are unlikely to provide ¾ private sector identity solutions will tend to be proprietary and will fail to capture the benefits of an open standard identity architecture as a public good. Thus, this approach fails to solve one of the problems which initially motivated government intervention.
Our discussion has focused on the U.S. government’s role in developing the global network. Because so much of Internet use is still concentrated in the U.S., the U.S. government perhaps has the leverage to dictate a standard. However, the Internet is a global network and thus it may suffer from a "race to the bottom" in which countries offering the greatest freedom and the least government regulation lure Internet companies and Internet servers. It is imperative that the U.S. government set a reasonable standard, if it decides to influence this marketplace, so that U.S.-based Internet companies do not have incentive to operate outside the country.
The most critical legal obstacle to the development and adoption of any effective digital identity mechanism is the current confusion over legal liability rules. In other words, who is responsible if someone’s digital identity is misused or stolen? Who bears the cost if a digital identification mechanism is compromised? The lack of a clear legal liability regime for these two issues discourages the cyberspace identity market from emerging in the first place and from operating efficiently once it does become widespread. Legislatures may need to enact liability laws that cover digital identity before the identity infrastructure can be effectively implemented.
The appropriate liability rules must reconcile two competing principles. First, because the market for digital identity mechanisms is in its infancy, the selected liability rules must help create incentives that will drive towards the widespread adoption of a secure identity infrastructure. According to this goal, the liability for identity misuse should be placed on whichever party can best induce the introduction and implementation of identity architecture. Second, in order to have an efficiently operating marketplace for identity mechanisms, it is desirable for the selected liability rules to place liability on the party who is the "least cost avoider" of harm. Adopting this goal, liability for identity misuse should be placed on whoever is best able to avoid misuse of digital identity. If these two goals point towards the same party, both goals can be accomplished together. However, if these two goals suggest that different parties should bear liability, then one goal or another must be made paramount, or the goals must be balanced.
In developing liability rules, the first situation we must consider is who bears the cost if digital identity is misused, stolen or forged in a cyberspace interaction. For example, suppose that person A is a valid, registered user of a Web site providing content, discussion groups, chat rooms and e-commerce. Suppose B, impersonating A, successfully logs on to a Web site as A, posts obnoxious messages, and orders merchandise using A’s pre-paid account on the Web site. Who should be liable for the harm caused by this fraudulent use of digital identity? In this context, there would be high transaction costs for all the parties facilitating a cyberspace interaction to negotiate ahead of time over liability, so the law should set a clear default rule.
We consider seven candidates for parties who could bear liability for digital identity misuse in cyberspace.
Victims: The person or entity whose identity is stolen or forged could be held liable. This would place on each person the burden of protecting his own identity. However, since users cannot directly dictate the security measures in use by their Web vendors, users would be forced, in effect, to ‘vote with their feet.’ Web services with better identity verification measures would flourish. Nonetheless, it would be difficult for many end-users to understand, let alone to judge, the effectiveness of identity mechanisms. Since the only sure way of protecting one’s identity in cyberspace would be to avoid cyberspace, users would curtail their cyberspace use. The chilling effect of this rule on the use of the Internet would be too significant.
Identity thieves: The person who stole or misused the victim’s identity could be held liable for the harms caused by his or her actions. This person, as the tortfeasor, clearly should be held at least jointly liable for the identity misuse. However, a rule placing exclusive liability on the identity thief will hold harmless the other parties who could and should have taken steps to protect the victim’s identity. This would lead to carelessness or inadequate measures to protect digital identity by the Web vendors. In addition, in the context of cyberspace, it is often very difficult to figure out exactly who the identity thief was, as by definition they were impersonating someone else. Further it may be difficult to find the individual even if his identity is known. If the identity thief alone were liable, then in most cases of identity misuse, no claims would ever be brought and no damages would ever be collected by the victims.
Application vendors: Software developers, such as those producing Web browsers or e-mail applications, could be held liable for any identity misuse that took place through their products. This type of rule would rapidly lead to the widespread implementation of identity mechanisms in all Internet software products. However, software products are general purpose tools. This rule would raise the cost of Internet software products for all users, as software vendors could not distinguish in advance between customers who would use the products for high-risk purposes and those who would use the products for low-risk purposes. Liability costs would be shifted to end-users, but in an unfair and inefficient manner under this rule.
Internet Service Providers: ISPs could be held liable for any identity misuse that is done to or by their clients. The concern with allocating liability in this way is that ISPs cannot know that an identity has been stolen unless the certificate authority or certificate holder notifies them (e.g., via a certificate revocation list or other authorized information). In addition, prices for Internet access for all users would rise because ISPs, like application vendors, cannot easily distinguish between low and high-risk Internet customers. In the situation where the ISP is notified of the revocation, it might be reasonable to assign some amount of liability if it negligently continues to permit continued exchange. Under this scenario, rather than raising the price of Internet access to all users, the ISP may contract with users in advance to include a penalty fee for sending a revoked certificate that exposes the ISP to liability.
Hardware vendors: In a non-trusted system, the hardware layer is quite distinct from the higher software layers, and misuse of one layer should not be the responsibility of those producing and implementing a different layer. Holding a hardware vendor liable for misuse of identity would place an undue burden on the manufacturer and would create a strong disincentive to develop hardware systems. In contrast, manufacturers of trusted systems might appropriately be held liable to some extent because the manufacturer can control the degree of security provided. Liability might be limited in cases where the public had been adequately notified of flaws and the manufacturer took appropriate steps to mitigate the impact of the problem.
Identity verifiers: Sources of trust, such as certificate authorities (CAs), could be held liable for the misuse of digital identity, if they were the least cost avoider. However, it is unlikely that CAs will be the least cost avoiders, as there is no way a CA could identify users whose identities are likely to be misused, or prevent their misuse. However, CAs do have a responsibility to verify the traits which they are certifying, and could reasonably be held liable in cases where certification is negligent.
CAs also should shoulder some liability for the adoption of digital identification systems that are inherently unreliable. Since liability cannot be placed directly on the developers of an open standard technology, we instead place liability on those who can pressure the developers to build a reliable system. The logical choice for the placement of liability is on the CA because CAs are in the business of providing trust, whereas other participants in the system simply utilize the infrastructure to achieve other ends. The drawback to this approach is that assigning liability to the CAs in this nascent market for secure identity mechanisms would require more reliable, hence more expensive systems. This might prevent the development of this market, prolong the implementation of an identity infrastructure, and dramatically raise the cost of entry for identity solutions developers. In order to mitigate the impact on CAs and increase the likelihood of market development, some other party involved in cyberspace interactions should also bear liability for identity misuse. This other party should be chosen to ensure that the CA and the other party can exert mutual pressure toward creating an identity system that has the fewest liability concerns.
Internet host servers: Internet hosts, such as on-line vendors or Web site or chat room providers, could be held liable for any identity misuse that takes place through their servers. For example, if person B successfully impersonates person A on Amazon.com, then Amazon.com would be liable for the damages. This rule places the liability on the party most able to develop or purchase identity mechanisms to prevent identity fraud because the owner of the Internet host is best able to install an identity verification system on its servers. This rule will create strong incentives leading to the deployment of identity verification mechanisms. However, this rule may promote systems that are highly specialized for each particular Internet host, rather than an open standard identity infrastructure. There would be an incentive to build a strong system for the lowest cost, but there would be no incentive to make the identity infrastructure interoperable if interoperability added any cost whatsoever to the host’s security bill.
Holding host servers strictly liable for any harm from identity misuse that takes place on their servers places the absolute burden of accurate identity verification on the party actually doing the verification. This rule does not require each Web host actually to build the identity mechanism or to bear the ultimate liability for identity misuse. Rather, this rule gives each Web host the appropriate incentive to make sure that all clients who access their servers are utilizing a reliable system of identification. Each Web host is in the best position to negotiate with other parties and to contract around this default liability rule to achieve for each particular situation a more efficient rule, if there is one. For example, Amazon.com might decide that VeriSign’s certificate procedures meet their needs for reliable verification; they might then form a contract indicating that in return for exclusively using VeriSign certificates, VeriSign would accept liability for abuse or fraud that results from use of their certificates. Alternatively, VeriSign might lure Amazon into such an exclusive contract by convincing Amazon of the integrity of VeriSign certificates and accepting some or all of the liability for misuse. This process need not be limited to a single certificate authority; indeed, one of the advantages of an open system is that Amazon could contract with multiple certificate authorities that it decided were trustworthy.
An additional element to the default liability rule may be necessary to promote the implementation of an open standard identity architecture. If one CA or one identity system was the market standard, then most Web hosts would probably contract with this CA. However, until and unless there is one predominant identity verification mechanism, each Web host would have the incentive to contract with the cheapest available CA who would agree to indemnify the Web host for any liability for misused identity. This is close to what is happening presently in the Internet marketplace: each Web host tends to use a separate identification scheme for its own purposes. Interoperability is basically nonexistent.
Thus, the strict liability rule could provide a form of "safe harbor" to promote the implementation of a standardized identity verification mechanism. There are several forms this safe harbor could take. At its most extreme, the law could disallow contracting and prohibit indemnification of liability unless the Web host contracts with a CA providing an interoperable identify architecture that conforms with standards set by the government or by an industry association. However, this type of rule is not feasible until there is an acceptable industry standard to mandate. A more flexible safe harbor provision might require that a Web host retain liability if it is negligent in selecting the identity infrastructure to adopt. This at least would require that Web hosts choose the identity architectures carefully and might weed out insecure mechanisms from the marketplace. Ultimately, the problem of developing an interoperable identity system may not be solvable merely by the creation of the correct liability rules. This problem might require market leadership by one company after a competitive period or might require regulatory intervention of some form from the government.
Assuming that a strict liability rule for Web hosts is adopted, and assuming further that the Web host has a contract with all trusted certificate providers, another liability issue remains: should the "identity issuer" (the CA who issues the identity certificate) or the "identity holder" (the user who receives and uses the identity certificate) be liable if the certificate is compromised? Until a rule for liability in this situation is determined, the marketplace for digital identity systems cannot operate efficiently, even if an identity infrastructure is implemented. Some method of assigning liability must therefore be determined.
There is certain to be a contractual relationship between the CA and the CA’s users — they are voluntarily in a relationship and they need to interact when the certificate is issued and the user’s identity is verified. No matter what default liability rule is set by law for this context, the contract between the two parties can and will reallocate this liability, unless this reallocation is made illegal. This contract will allocate liability in whole or in part to achieve the optimal allocation, unless for some reason the CA has inappropriate bargaining power over the user.
Whether it is optimal for the CA or the user to bear liability for a compromised certificate depends in part on which party is the "least cost avoider" of compromised certificates. Whichever party can most easily avoid the loss of the identity certificate should bear the liability and thus have incentive to take appropriate care in protecting the certificate. Who this party is depends on the technology that is used, on what physical manifestation the identification certificate involves, and on how difficult it is for the user to share the certificate inappropriately with others. If the certificate is something that can be shared easily by the user, then the user is the least cost avoider. For example, passwords can rapidly and simply be distributed to other people, so it would be inefficient to make providers of password-based identification mechanisms liable for compromised passwords. The users of the passwords must bear the liability for loss so that they have incentive not to share the passwords with others. For digital certificates such as under our proposal, the efficient allocation of liability between CA and user depends on how and where the certificate can be stored. If it is stored in a file on the user’s computer and its use is not linked in any way to the user’s computer, then it is effectively very similar to a password and the liability for loss or compromise should remain with the user. However, for more difficult-to-distribute certificates — such as those that are stored on smartcards or held in an encrypted fashion —liability for loss or compromise should fall on the CA. Biometric verification methods, where the certificate is in some way the user’s own body, are no doubt the hardest for the user to share inappropriately, so liability for compromise of these should also remain with the CA. A biometric-based certificate should only be compromised if the issuer’s technology fails.
Even in situations where the user is liable for a lost or compromised password or certificate, the user’s liability should be based on a standard of negligence. If the user takes reasonable care of his or her certificate and it is compromised nonetheless, then the user should not be liable. Defining exactly what is "reasonable care" of a digital certificate may be a difficult process. Consumer protection laws may be required if CAs try to allocate by contract liability even in cases where the user was not negligent or where the CA is the least cost avoider. For example, if the CA’s identification system is at fault for the misuse of identity (e.g., the system erroneously verifies an out-of-date password), then the CA should be strictly liable, regardless of the user’s possible negligence.
The global nature of the Internet adds a further level of complexity to any identity liability regime. What we have been considering so far is liability rules that would apply within one nation. If identity is misused across international borders, as is easily the case in many cyberspace interactions that involve users and servers and companies in several different countries, the liability rules become far more complex and are therefore beyond the scope of this paper. Which nations’ laws should apply? What if one nation has much more protective liability rules than another? What if the different national rules make different parties responsible? Which country should have legal jurisdiction? While, theoretically, the simplest solution would be to develop global legal rules that govern the Internet, this is not a realistic possibility, as many countries have very different concerns for privacy and freedom and different schemes for handling regulation.
The digital identity laws passed or under consideration even in the United States so far are controversial. It is difficult for lawmakers to understand and respond to the rapidly changing issues in cyberspace. Currently, many conflicting legislative efforts are underway, from those placing strict liability on the certificate authority to those that rely very heavily upon existing contract law and the courts to determine liability. As we consider the construction of a digital identity system, we need to consider the appropriate allocation of liability among these various parties.
Even without legislatures enacting new laws for digital identity liability, there are less formal methods by which to encourage the adoption and implementation of cyberspace identity verification mechanisms. Faster moving groups than state legislatures could pass rules or guidelines that encourage the use of digital identity mechanisms in cyberspace. For example, state bar associations that propose guidelines for lawyers’ conduct could be a point of significant leverage. Bar Associations could try to mandate that lawyer-client e-mail correspondence is not protected by the lawyer-client privilege unless the e-mail is encrypted. In the real world, the lawyer-client privilege only applies to information that is kept in confidence by the lawyer and the client, not to information knowingly revealed to third parties; in cyberspace this type of rule would make sense, since non-encrypted e-mail is basically open to any third party’s observation and thus, arguably, "knowingly revealed." Setting a rule like this would force all lawyers to use encryption mechanisms in cyberspace, and all clients would probably have to follow suit, potentially spurring the need for a more standardized digital identity architecture. Only if a group leads the way and demonstrates the need for and use of a digital identification architecture will such a system become widely adopted.
We have outlined the functionality and basic operation of a secure, open standard for a digital identity verification architecture. Our system would resolve many of the problems with existing identity mechanisms and could be implemented to produce a much more secure cyberspace environment. Our system is flexible enough that it could be used to meet the requirements in a wide variety of settings, from social to business to government. Exactly how the certificates and methods that we have described should best be implemented in the marketplace depends on the needs of the particular identity application and the choice of technology.
In broad terms, there are just three types of identification mechanisms. Authentication can be based on: a person’s shared knowledge (such as a password); a person’s possession of unique information or device (such as a digital certificate); or a person’s inherent unique characteristics (such as a fingerprint or other biometric). While our system fits mostly in the second category, it would be perfectly compatible with the other two approaches. If, in a certain context, a very strong link to real world persona is needed, then a biometric system could be the "front-end" for our architecture. Likewise, a password or PIN could be required for access to a certificate to prevent simple copying of identity certificates.
While there is flexibility about exactly how our identity architecture should be implemented, the technology now exists to make our proposal work. The choices are just exactly how, and where, and by whom it should be built. Cyberspace will be a better, safer, and more useful place when people know for sure that they are communicating with a dog.
Appendix 1: Legal Process For Law Enforcement Access To The Trace
Congress will have to develop statutes to govern the conditions under which law enforcement agencies can access the trace feature, should such a feature be included in the digital identity architecture.
It is important to understand the exact type of information to which law enforcement will have access under the trace feature. Consider the example transaction discussed in an earlier section of this paper. In that transaction, Bob, using the alias "Ghost," purchased a product that required him to prove that he was over the age of eighteen. Under the trace, law enforcement would contact the web site store and ask for the alias of the person who made the transaction, and would then contact the certificate authority and ask for the identity of "Ghost." In legal circles, this identity information is referred to as transactional information, to distinguish it from content information. Law enforcement would be requesting content information if they asked the web site store what "Ghost" told the web site about himself — that he is over eighteen, in our example. Typically, content information is afforded higher procedural protections than transactional information.
Congress could rely on a number of current statutes as a model for new legislation that would govern law enforcement access to the trace, however, none of the current statutes provides a perfect model. This section will explore some of the features Congress may wish to borrow from two of these statutes, the Wiretap Statute (Title III) and the Stored Electronic Communications Privacy Act (ECPA).
Although Title III is designed to regulate government interceptions of communications (content information), the statute does contain some features that Congress may wish to apply to the access provisions for the law enforcement trace. First, Title III requires law enforcement to obtain a court order. Although current laws sometimes allow law enforcement to access transactional information without first obtaining a court order or warrant, Congress may wish to implement a court order requirement because of the current political sensitivity regarding privacy on the Internet. Second, Title III requires law enforcement to show that traditional means have failed or would fail before a wiretap authorization can be granted. Congress may wish to consider whether there might be some other method by which law enforcement could obtain this information and whether there should be a statutory preference for using that method over the law enforcement trace. Finally, Title III requires law enforcement to minimize the interception so communications not covered by the order authorizing the wiretap are not intercepted accidentally. Following this principle, Congress should draft legislation that prohibits law enforcement from accessing anything other than transactional logs, particularly archived copies of exchanged digital certificates.
Another statute that Congress should review when drafting legislation to regulate access to the law enforcement trace is ECPA. ECPA’s provisions establish procedures for governmental access to both content and transactional information stored with electronic communication providers. As ECPA does for electronic communication providers, Congress will want to prohibit certificate authorities and web site stores from disclosing identity information and related transactional information unless consent from the appropriate individual is obtained or the request comes from an authorized law enforcement officer. Although ECPA allows for access to certain types of information with an administrative subpoena, as mentioned above, for political reasons, Congress may find it necessary to require a warrant for access to the trace. One interesting twist of ECPA is that it places liability on the electronic service provider for disclosing information to a government agent who has not followed the statute’s procedures, but does not place any liability on the government for requesting information in a manner that does not conform to the access procedures established in ECPA. Regardless of whether this was an intentional decision by Congress or an oversight, Congress should require the government to follow process in any new statute governing a law enforcement trace because such a law would offer an individual the possibility that improperly obtained information could be deemed inadmissible and excluded from any criminal proceedings.
Appendix 2: Digital Signature Legislation
For purposes of this discussion, it is important to distinguish between digital and electronic signatures. A digital signature is "an electronic identifier that utilizes an information security measure, most commonly cryptography, to ensure the integrity, authenticity, and nonrepudiation of the information to which it corresponds." In contrast, an electronic signature refers to "any identifiers such as letter, characters, or symbols, manifested by electronic or similar means, executed or adopted by a party to a transaction with an intent to authenticate a writing."
In the United States, individual states have led the way in the passage of legislation regulating the use of digital and/or electronic signatures. State digital and/or electronic signature legislation generally can be categorized as one of three types of legislation: prescriptive, criteria-based, or signature-enabling. Legislation following the prescriptive model provides a specific regulatory and statutory framework for the recognition of digital signatures. Criteria-based legislation requires signatures to satisfy certain criteria of reliability and security in order to be legally binding. Finally, signature-enabling legislation, such as that passed in Florida and Massachusetts, permits any electronic mark that is intended to authenticate a writing to satisfy a signature requirement.
In 1991, the American Bar Association’s Information Security Committee began to draft a model law for digital signatures. Four years later, there was still significant disagreement within the Committee over key components of the model legislation, so, in the summer of 1995, the Committee released its work in the form of guidelines. The ABA’s Digital Signature Guidelines are prescriptive—they attempt "to delineate a comprehensive scheme for the recognition of digital signatures in a PKI environment utilizing state-licensed certification authorities." These prescriptive guidelines became the basis for much state digital signature legislation.
Utah’s digital signature law, passed in 1995 and amended in 1996, was the first of its kind in the United States. Influenced by the ABA’s efforts and enacted just before the ABA’s Guidelines were released, Utah’s law followed the prescriptive model by providing a specific regulatory and statutory framework for the recognition of digital signatures. Indeed, Utah's digital signature statute epitomizes prescriptive, PKI-based legislation: it "establishes a detailed PKI licensing scheme, allocates duties between contracting parties, prescribes liability standards, and creates evidentiary presumptions and standards for signature or document authentication." Under the Utah Act, a digitally signed document satisfies the writing requirements if the signature is verified by a valid licensed public key, however, only attorneys, financial institutions, title insurance companies, and the State of Utah may act as licensed certification authorities, and licensed Certification Authorities are required to post a guaranty in the form of a bond or letter of credit. The Utah Act does not prohibit the operation of unlicensed certification authorities in Utah, however, unlicensed CAs lose evidentiary presumptions of authenticity and do not enjoy the benefit of limited liability under the Act. Note the potential effect of such laws on CAs wishing to certify "low-value" certificates. In order to enjoy evidentiary presumptions of authenticity an limited liability, these CAs need to be licensed; however the costs associated with such licensing could be prohibitive for CAs distributing "low-value" certificates.
Following the lead of the ABA Information Security Committee and the Utah legislature, some states have adopted prescriptive legislation, however other states, mindful of the drawbacks of the prescriptive approach, have eschewed this technology-specific approach in favor of a more flexible approach. The prescriptive approach limits liability and establishes evidentiary presumptions when users rely on digital signature technology used in conjunction with state-licensed certification authorities. Simultaneously, such legislation discourages reliance on alternatives, even if they superior security, by denying them the benefit of these presumptions. Thus, some states have opted for more flexible legislation that is technology-neutral and avoids "market-distorting effects" by choosing not to define a particular liability regime.
Other states have adopted criteria-based legislation that requires signatures to satisfy certain criteria of reliability and security in order to be legally binding. For example, under California's criteria-based law, an electronic signature is legally effective if it satisfies the following criteria: it is unique to the person using it; it is capable of verification; it is under the sole control of the person using it; it is linked to the data in such a manner that if the data is changed the signature is invalidated; and it is in conformity with regulations adopted by the appropriate state agency (the Secretary of State).
Finally, signature-enabling legislation, such as that passed in Florida and Massachusetts, permits any electronic mark that is intended to authenticate a writing to satisfy a signature requirement. For example, the Florida Electronic Signature Act of 1996 provides that, unless otherwise provided by law, an electronic signature may be used to sign a writing and shall have the same force and effect as a written signature. Under the statute, "electronic signature means any letters, characters, or symbols, manifested by electronic or similar means, executed or adopted by a party with an intent to authenticate a writing," and a writing "is electronically signed if an electronic signature is logically associated with such writing."
States are now experimenting with "hybrid" digital signature legislation that combines aspects of two or more of the above approaches. The recent trend in state digital signature legislation has been toward legislation that not only removes barriers to electronic commerce, but affirmatively enables electronic commerce and establishes evidentiary presumptions in favor of the electronic signature user based on security and trustworthiness standards.
The NCCUSL is working to develop uniform state laws of electronic commerce. Two major NCCUSL projects of interest include the draft revision of Article 2B of the Uniform Commercial Code (expected to be approved in July of 1999 and enacted by the states in 2000), and the March 23, 1998 draft Uniform Electronic Transactions Act (UETA) (scheduled for final approval in the summer of 1999).
The Clinton Administration's "Framework for Global Electronic Commerce," released July 1, 1997, emphasized that electronic economic activity currently is conducted in an atmosphere of legal uncertainty. In particular, the report highlighted existing uncertainties regarding the validity of electronic records and documents used to evidence commercial transactions and relationships. To address these legal uncertainties, the administration's Framework called for the development of a Uniform Commercial Code for Electronic Commerce. In response, the NCCUSL began drafting the Uniform Electronic Transactions Act (UETA) which is intended to create a basic legal structure recognizing and effectuating electronically generated records and signatures. The UETA is scheduled for final NCCUSL approval in the summer of 1999.
The goal of the committee working on the UETA is to "create a law that will be adopted by the 50 states to govern several key aspects of electronic commerce," however, there recently has been considerable debate over the appropriate scope of the UETA. In particular, the UETA Drafting Committee was struggling to decide whether the Act merely should remove existing barriers to electronic commerce or should go further and actively promote electronic commerce. There was agreement that the UETA should remove existing barriers to electronic commerce, including uncertainty over whether or not electronic messages are "legal" [i.e., satisfy statutes and regulations requiring transactions to be "in writing" and "signed"] and whether or not electronic records are admissible. However, there was disagreement over whether or not the UETA should establish legal presumptions (e.g., the law would presume the authenticity and integrity of a message if certain types of security procedures were implemented) in order to facilitate reliance on electronic messages. The Drafting Committee recently voted to remove all references to presumptions in the UETA, reflecting a focus on the removal of existing barriers to electronic commerce, rather than on affirmative facilitation of electronic commerce.
Several provisions of the draft law reflect the idea that electronic media should be treated on a par with written media. One such provision holds that "[a] record may not be denied legal effect, validity, or enforceability solely because it is an electronic record." Indeed, where "existing law requires a record be in writing in order to be enforceable, the UETA simply provides that the existence of an electronic record satisfies that requirement."
Similarly, the NCCUSL draft codifies the fundamental premise of the Act that "the form in which a signature is generated, presented, communicated or stored may not be the only reason to deny the signature legal recognition." Indeed, a provision of the UETA ensures that "[a] signature may not be denied legal effect, validity, or enforceability solely because it is an electronic signature." Unlike certain existing state laws, the UETA does not set forth substantive requirements for the creation of valid electronic signatures. For purposes of the act, however, "an electronic record will be deemed signed by an electronic signature if the signature is ‘verified in conformity with a commercially reasonable security procedure.’" While this technology-neutral approach will foster competition by allowing a variety of security measures to blossom; there is likely to be uncertainty, at least initially, in determining what meets the "commercially reasonable" standard (i.e., what constitutes "commercially reasonable" security).
The draft law also creates presumptions regarding the identity and integrity of electronic records and signatures where heightened security procedures are followed.
Regulation of digital and/or electronic signatures mostly has taken place at the state level. The recent NCCUSL drafting efforts likely reflect a perceived need for greater uniformity among states in the treatment of digital and/or electronic signatures. Yet, some view these efforts as inadequate to achieve the necessary uniformity, and instead favor federal legislation.
Rationales posited in support of federal statutory intervention include:
There are several bills pending in the U.S. Congress that would provide for uniform national treatment of digital and/or electronic signatures. These bills are listed below and briefly summarized in the attached table.
Title, Citation, Sponsor, and Status of Legislation
Government Paperwork Elimination Act
Electronic Commerce Enhancement Act
Digital Signature and Electronic Authentication Law (SEAL) of 1998
Computer Security Enhancement Act of 1997
Electronic Financial Services Efficiency Act of 1997
American Bar Association (ABA) CyberNotary Committee Homepage
Digital Signature Advisory Committee of the Florida Department of State, Electronic Commerce in Florida: Report to the Florida Legislature's Joint Committee on Information Technology Resources, November 30, 1996.
Digital Signature Initiatives-A Summary
Digital Signature Links
Federal Digital Signature Legislative Initiatives-A Summary
Federal PKI Task Force of the Government Information Technology Services Board (GITSB), Model Certificate Policy
National Conference of Commissioners on Uniform State Laws Drafts of Uniform and Model Acts
Resources and Legislation
Pending Federal Initiatives
National Conference of Commissioners on Uniform State Law (NCCUSL) Drafting Efforts