If you compare the digital interceptor, to the voice interceptor, fairly, you will see they are in equally strong positions. When I am phoneing a person I know, I am automatically checking the `signature' of their voice. The other party on the line might be able to convince me they have a cold, but I hope I will have enough wisdom to postpone discussing the March 15th assassination plot untill the cold clears up. So we should compare a voice interceptor on a channel where the two people don't know each other's voice to the unsigned digital interceptor. In this case, the interceptor can claim to one party to be the other party, and remain undetected. This is the Diffe-Helman weakness. Alternatively we should compare the voice interceptor on a channel where the two people do know each other's voice to the signed digital interceptor. In this case, the interceptor will either be detected should some minimal authentication and verification be tried, or the interceptor will be unable to even listen in. The weakness remains here, but it has been patched over with authentication, and signed verification of the channel key. This is the Diffe-Helman weakness weakness. The (potential) interceptor is the reason why we must be so very carefull when validating other people's public keys. I know there is no interceptor between me and the people who's keys I sign. If I can be sure of no interceptors between one of them, and the person I wish to speak to, then I will be able to establish a secure channel. The AT&T Telephone Security Device (you know, the beast with the Clipper chip...) has a display that shows a few digits of a hash of the key. Each party reads off some of it to the other; the idea is that an intruder won't be able to spoof a voice in real-time. For data connections, have a look at @article{interlock, author = {Ronald L. Rivest and Adi Shamir}, journal = {Communications of the ACM}, number = {4}, pages = {393--395}, title = {How to Expose an Eavesdropper}, volume = {27}, year = {1984} } The idea is to send half of an encrypted block. You then await a half-block from the far side, at which point you send your other half, and listen for the far side's other half. The idea is that since one can't decrypt a half-block, the intruder in the middle can't send a fraudulent one. Depending on how this scheme (known as the ``Interlock Protocol'') is used, it may be vulnerable to attack. Davies and Price, in their (excellent) book ``Security for Computer Networks'', suggest sending passwords that way. But Mike Merritt and I showed how to attack that scheme under certain circumstances. (Details to appear in IEEE Transactions on Information Theory.) --Steve Bellovin