Anthropic says its Mythos model found vulnerabilities in cryptographic algorithms that secure the internet

Anthropic says Claude Mythos Preview found an improved attack on the HAWK post-quantum signature candidate and a new attack on a reduced-round AES-128 variant.

The HAWK result reportedly took about 60 hours and around $100,000 in API costs; the AES work ran mostly autonomously over multiple days.

Anthropic says the vulnerabilities do not affect currently deployed systems and has shared the findings with government and industry partners.

The company also launched CryptanalysisBench, a benchmark for evaluating language models on cryptanalysis tasks.