[empath75]

Since it seems that both adversaries in the network are training in parallel, is it possible that the encryption is only exploiting a weakness in that particular Eve? Would it change anything to have more Eves challenging Alice and Bob?

Also-- being able to generate crypto-algorithms on the fly seems like it would be ideal for small cells of people who want to keep their communications secret from something like the NSA, who might be looking for something like RSA or GPG, but not some ai generated by a neural network that nobody else in the world is using.

Oh- and how susceptible is the generated ciphertext to standard cryptographic techniques like letter frequency analysis and so on.

[dgacmu]

Yes. Some of this is in the paper, but I didn't try training with multiple eves at a time (yet). It's a very reasonable thing to try. We did test the robustness by doing a final evaluation pass where we trained a fresh eve a few dozen times without modifying A&B. That eve was generally 0.5-2 bits more successful than the one being trained iteratively, suggesting we could do better.

The last question you asked is, well, a good question. There's no reason to think that the current algorithm is very good in that regard. It's probably vulnerable, since we know it mixes multiple key bits & plaintext bits together.