TLDR Tech / InfoSec
Some Thoughts About Anthropic's New Cryptanalysis Results
Anthropic recently announced that its unreleased Mythos model discovered two novel attacks against the post-quantum cryptography hashing candidate HAWK and a reduced-round AES variant. The HAWK attack was in fact novel, but it only effectively reduced the strength of the cipher by halving the bits, whereas the AES attack was a slight speed-up of a theoretical attack from 2013. The model was able to develop the attacks with a naive prompt, and from that perspective, the results are still meaningful.
Read the full story