Cryptographer Analyzes Anthropic's AI Cryptanalysis Results Beyond the Hype

JeremyCMorgan · x · 2026-08-06

Cryptographer Matthew Green provides a sober analysis of Anthropic's recent cryptanalysis results generated by their unreleased model, Claude Mythos. The AI successfully developed a key recovery attack against the post-quantum signature scheme HAWK and improved attacks on reduced-round AES.

Green clarifies the practical implications: HAWK is still in the evaluation phase and not widely deployed, meaning the attack doesn't break active systems or transfer to the standardized Falcon scheme. He emphasizes that while AI-discovered vulnerabilities are particularly relevant during the current post-quantum migration, it's crucial to separate genuine technical substance from media sensationalism.

Original post →

More from Safety

Safety channel →