Discovering cryptographic weaknesses with Claude

Anthropic describes work by its Claude Mythos system that autonomously discovered and improved cryptanalytic attacks: a key‑recovery attack that reduces the effective key strength of the HAWK post‑quantum signature candidate, and a faster meet‑in‑the‑middle attack (the “Möbius Bridge”) on a reduced‑round AES variant, along with other research on LEA and Serpent reduced‑round attacks; the findings are presented as research results with coordinated disclosure and are asserted to have no immediate practical impact on deployed systems.