Anthropic says its Mythos model found post-quantum crypto weaknesses

Anthropic says its Claude Mythos Preview model found weaknesses in important cryptographic algorithms, including a stronger attack on HAWK, a post-quantum signature scheme.

The Decoder reports that human experts had reviewed HAWK for more than two years, while the model found the improved attack in about 60 hours at an API cost of roughly $100,000. Anthropic says the findings do not affect systems currently in use.

The result is not an internet-breaking event, but it is a warning about the pace of AI-assisted analysis. Cryptography depends on deep review by specialists, and models that can search proof spaces or attack designs could change how quickly assumptions are tested.

The practical implication is double-edged. AI may help defenders find flaws earlier, before standards are widely deployed. It may also give attackers more capability. Cryptographic review may need to include AI systems as regular tools, not occasional experiments.