Anthropic's Mythos Model Finds Practical Attacks on HAWK-256 and Reduced-Round AES

An unreleased Claude variant reportedly discovered new cryptographic attacks, turning a frontier model into a machine-speed cryptanalysis engine.

Anthropic's unreleased Mythos model has reportedly discovered new cryptographic attacks, including practical attacks on the HAWK-256 scheme and reduced-round AES, according to @ivke2006. If the results hold up under scrutiny, it is one of the more concrete demonstrations to date that frontier models can do original security research rather than merely summarize it.

Unlock the full briefing

Get every story in today's briefing, the full archive, and the daily AI intelligence brief.

All stories today

Full archive

Daily brief

Cancel anytime. Payments powered by Stripe.