Anthropic's Claude Mythos Preview discovers an attack technique that halves post-quantum cryptographic strength

Anthropic says Claude Mythos Preview found a stronger attack on HAWK, a post-quantum signature scheme, effectively halving its key strength.

The model also improved an attack on a weakened AES variant, making it hundreds of times faster.

The findings show AI models could become powerful tools for cryptographic research and security review.

At the same time, they raise concerns that AI may speed up discovery of weaknesses in widely used systems.