Here’s what Anthropic found when it turned Mythos loose on encryption algorithms

Anthropic used its Claude Mythos AI to find cryptanalytic weaknesses in two areas: a lattice automorphism that halves the effective key strength of HAWK (a NIST post‑quantum signature candidate), and a theoretical “Möbius Bridge” shortcut that makes the strongest known attack on a seven‑round AES test variant hundreds of times faster; neither flaw impacts deployed full‑strength systems, the AES result is impractical due to impossible data requirements, and Anthropic coordinated disclosure and published tools to let other researchers reproduce the work.