Anthropic says Claude found new attacks on HAWK and round-reduced AES

jedisct1 · x · 2026-07-29

Using Claude Mythos Preview, Anthropic researchers found two new cryptanalytic attack paths:

Anthropic says these are substantial research results, but they do not affect production systems today. The post frames the work as evidence that frontier models can help uncover weaknesses not just in implementations, but in the mathematical structure of cryptographic algorithms themselves.

Related event: Anthropic's Claude Mythos Uncovers Flaws in Post-Quantum Crypto and AES(30 posts)→

Original post →

More from Research

Research channel →