Untraceable Electronic Mail, Return addresses, and Digital Pseudonyms David L. Chaum University of California, Berkeley Abstract A technique based on public key cryptography is presented that allows an electronic mail system to hide who a participant communicates with as well as the content of the communication - in spite of an unsecured underlying telecommunication system. The technique does not require a universally trusted authority. One correspondent can remain anonymous to a second, while allowing the second to respond via an untraceable return address. Key Words and Phrases: electronic mail, public key cryptosystems, digital signatures, traffic analysis, security, privacy Introduction Cryptology is the science of secret communication. Cryptographic techniques have been providing secrecy of message content for thousands of years [3]. Another cryptographic problem, "the traffic analysis problem" (the problem of keeping confidential who converses with whom, and when they converse), will become increasingly important with the growth of electronic mail. This paper presents a solution to the traffic analysis problem that is based on public key cryptography. The following two sections introduce the notation and assumptions. Then the basic concepts are introduced for some special cases involving a series of one or more authorities. The final section covers general purpose mail networks. Notation Someone becomes a user of a public key cryptosystem (like that of Rivest, Shamir, and Adleman [5]) by creating a pair of keys K and Inv(K) from a suitable randomly generated seed. The public key K is made known to the other users or anyone else who cares to know it; the private key Inv(K) is never divulged. The encryption of X with key K will be denoted K( X ), and is just the image of X under the mapping implemented by the cryptographic algorithm using key K. The increased utility of these algorithms over conventional algorithms results because the two keys are inverses of each other, in the sense that Inv(K)( K( X ) ) = K( Inv(K)( X ) ) = X. A message X is "sealed" with a public key K so that only the holder of the private key Inv(K) can discover its content. If X is simply encrypted with K, then anyone could verify a guess that Y = X by checking whether K( Y ) = K( X ). This threat can be eliminated by attaching a large string of random bits R to X before encrypting. The sealing of X with K is then denoted K( R, X ). A user "signs" some material X by prepending a large constant C (all zeros, for example) and then encrypting with its private key, denoted Inv(K)( C, X ) = Y. Anyone can verify that Y has been signed by the holder of Inv(K) and determine the signed matter X, by forming K( Y ) = C, X and checking for C. Assumptions The approach taken here is based on two important assumptions: (1) No one can determine anything about the correspondences between a set of sealed items and the corresponding set of unsealed items, or create forgeries without the appropriate random string or private key. (2) Anyone may learn the origin, destination(s), and representation of all messages in the underlying telecommunication system and anyone may inject, remove, or modify messages. Mail System The users of the cryptosystem will include not only the correspondents but a computer called a "mix" that will process each item of mail before it is delivered. A participant prepares a message M for delivery to a participant at address A by sealing it with the addressee's public key Ka, appending the address A, and then sealing the result with the mix's public key K1. The left-hand side of the following expression denotes this item which is input to the mix: K1( R1, Ka( R0, M ), A ) --> Ka( R0, M ), A. The --> denotes the transformation of the input by the mix into the output shown on the right-hand side. The mix decrypts its input with its private key, throws away the random string R1, and outputs the remainder. One might imagine a mechanism that forwards the sealed messages Ka( R0, M ) of the output to the addressees who then decrypt them with their own private keys. The purpose of a mix is to hide the correspondences between the items in its input and those in its output. The order of arrival is hidden by outputting the uniformly sized itemsin lexicographically ordered batches. By assumption (1) above, there need be no concern about a cryptoanalytic attack yielding the correspondence between the sealed items of a mix's input and its unsealed output - if items are not repeated. However, if just one item is repeated in the input and allowed to be repeated in the output, then the correspondence is revealed for that item. Thus, an important function of a mix is to ensure that no item is processed more than once. This function can be readily achieved by a mix for a particular batch by removing redundant copies before outputting the batch. If a single mix is used for multiple batches, then one way that repeats across batches can be detected is for the mix to maintain a record of items used in previous batches. (Records can be discarded once a mix changes its public key by, for example, announcing the new key in a statement signed with its old private key.) A mix need not retain previous batches if part of each random string R1 constains something - such as a time-stamp - that is only valid for a particular batch. If a participant gets signed receipts for messages it submits to a mix, then the participant can provide substantial evidence that the mix failed to output an item properly. Only a wronged participant can supply the receipt Y (= Inv(K1)( C, K1( R1, Ka( R0, M ), A ))), the missing output X (= Ka( R0, M ), A ), and the retained string R1, such that K1( Y ) = C, K1( R1, X ). Becasue a mix will sign each output batch as a whole, the absence of an item X from a batch can be substantiated by a copy of the signed batch. The use of a "cascade", or series of mixes, offers the advantage that any single constituent mix is able to provide the secrecy of the correspondence between the inputs and the outputs of the entire cascade. Incrimination of a particular mix of a cascade that failed to properly process an item is accomplished as with a single mix, but only requires a receipt from the first mix of the cascade, since a mix can use the signed output of its predecessor to show the absence of an item from its own input. An item is prepared for a cascade of n mixes the same as for a single mix. It is then successively sealed for each succeeding mix: Kn( Rn, K( R, ... , K2( R2, K1( R1, Ka( R0, M ), A ))...)) -->. The fist mix yields a lexicographically ordered batch of items, each of the form K( R, ... , K2( R2, K1( R1, Ka( R0, M ), A ))...) -->. The items in the final output batch of a cascade are of the form Ka( R0, M ), A, the same as those of a single mix. Return Addresses The techniques just described allow participant x to send anonymous messages to participant y. What is needed now is a way for y to respond to x while still keeping the identity of x secret from y. A solution is for x to form an untraceable return address K1( R1, Ax), Kx, where Ax is its own real address, Kx is a public key chosen for the occasion, and R1 is a key that will also act as a random string for purposes of sealing. Then, x can send this return address to y as part of a message sent by the techniques already described. (In general, two participants can exchange return addresses through a chain of other participants, where at least one member of each adjacent pair knows the identity of the other member of the pair.) The following indicates how y uses this untraceable return address to form a response to x, via a new kind of mix: K1( R1, Ax ), Kx( R0, M ) --> Ax, R1( Kx( R0, M )). This mix uses the string of bits R1 that it finds after decrypting the address part K1( R1, Ax ) as a key to re-encrypt the message part Kx( R0, M ). Only the addressee x can decrypt the resulting output because x created both R1 and Kx. The mix must not allow address parts to be repeated - for the same reason that items of regular mail must not be repeated. This means that x must supply y with a return address for each item of mail x wishes to receive. Also notice that conventional as opposed to public key cryptography could be used for both encryptions of M. With a cascade of mixes, the message part is prepared the same as for a single mix, and the address part is as showin in the following input: K1( R1, K2( R2, ..., K( R, Kn( Rn, Ax ))...)), Kx( R0, M ) -->. The result of the first mix is K2( R2, ..., K( R, Kn( Rn, Ax ))...), R1( Kx( R0, M )) -->, and the final result of the remaining n-1 mixes is Ax, Rn( R ... R2( R1( Kx( R0, M )))...). Untraceable return addresses allow the possiblity of "certified" mail: They can provide the sender of an anonymous letter with a receipt attesting to the fact that the letter appeared intact in the final output batch. The address A that is incorporated into a certified letter is expanded to include not only the usual address of the recipient, but also an untraceable return address for the sender. When this return address appears inthe output batch of the final mix, it is used to mail the sender a signed receipt which inlucdes the message as well as the address to which it was delivered. The receipt might be signed by each mix. Summary and Conclusion A solution to the traffic analysis problem has been presented that allows any single intermediary to provide security for those messages passing through it. In addition, the solution allows messages to be sent or received pseudonymously. Through the notion of a roster of pseudonyms, it also provides some new and interesting kinds of limited anonymity. References 1. Baran, P. On distributed communications: IX security secrecy and tamper-free considerations. Memo RM-3765-PR, Rand Corp., Santa Monica, CA, Aug. 1964. 2. Diffie, W. and Hellman, M.E. New directions in cryptography. IEEE Trans. Information Theory IT-22, 6 (Nov. 1976), 644-654. 3. Kahn, D. The Code Breakers, The Story of Secret Writing. Macmillan, New York, 1967. 4. Merkle, R.C. Secure communications over insecure channels. Comm. ACM 21, 4 (Apr. 1978), 294-299. 5. Rivest, R.L., Shamir, A., and Adleman, L. A method for obtaining digital signatures and public-key cryptosystems. Comm. ACM 21, 2 (Feb. 1977), 120-126.