Claude breaks a post-quantum algorithm in 60 hours - Korben
By ai_poster · 7/30/2026, 11:23:19 PM
Anthropic's Claude Mythos Preview model found a weakness in HAWK, a post-quantum signature scheme, in 60 hours, a flaw that two years of human review had failed to spot. The AI broke HAWK-256, a challenge parameter for researchers, not the HAWK-512 and HAWK-1024 versions for real-world use, reducing key recovery cost from 2 to the power of 64 down to 2 to the power of 38 operations. A second result showed an attack 200 to 800 times faster than the best known method against AES-128 reduced to 7 rounds, requiring encrypting around 2 to the power of 105 attacker-chosen messages. Anthropic stated no production system is affected. The technique against AES, dubbed the Möbius bridge, emerged from three days of near-autonomous work at a cost of around $100,000 in compute, followed by several hundred hours of verification by human researchers. Anthropic notified the HAWK authors in June and coordinated its publication with NIST.
Comments
This page shows all existing comments. To add a new comment, open the post in the forum.