There has been considerable discussion on sci.crypt and on the cypherpunks list about the fact that currently proposed digital cash is "traceable", or to put it another way, that there is no payee anonymity. This is an annoying asymmetry, where the payor is protected more than the payee. But there is a fundamental reason for this, which I want to explain here. It is not just perversity on the part of digital cash designers. The problem is that there is a conflict between the desire for payee anonymity and the need to prevent double spending. And preventing double spending is far more important, since without that the cash would be worthless. Here is how the conflict occurs. Suppose Alice has a piece of digital cash which she wants to spend with Bob. She goes through some protocol and transfers data to him. Bob, then or later, sends some resulting data to the bank and gets his account credited. Now if Alice spent that same coin with Charlie, we need to have the bank find it out. When Charlie deposits his data with the bank, and the bank compares that with what Bob sent in, there must be a red flag that goes up. The fundamental requirement of preventing double spending implies that Bob's and Charlie's data, when sent to the bank, has some correlation which will identify the fact that they both come from the same coin. It doesn't matter exactly what the form of this data is, or how it has been blinded and stirred, but if double spending is to be detected there must be a correlation which the bank can see. But this correlation is what makes the coin traceable. Suppose Alice is paying a coin to Bob via an anonymous network, and she and the bank are going to try to figure out who he really is. She goes through the payment transaction, and Bob sends his resulting data to the bank. Before doing so, though, Alice simulates a payment of the same coin to Charlie. Charlie doesn't actually have to be involved, Alice can just go through what she would have done if she had spent the coin elsewhere. The result of this simulated payment has been shared with the bank. Now, when Bob deposits his data, the bank compares it with the data Alice sent, the result of her simulated spending of the same coin. By the argument presented above, Bob's deposit will be flagged. It will correlate with the data Alice sent in since this will be the equivalent of a double-spending. So when Bob makes the deposit he can be linked to the specific coin payment which Alice made, and his anonymity is lost. It would seem that any system which is capable of detecting double- spending just from the information which the payees send in to the bank would be vulnerable to this. Systems which use tamper-proof observer chips to prevent double spending beforehand can avoid it, but of course if someone breaks an observer the whole cash system might crash. In general it does not look like payee anonymity is possible without giving up other very important features. Hal Finney hfinney@shell.portal.com