Anthropic’s Claude AI cracks weaknesses in post-quantum digital signature scheme in 60 hours

Anthropic’s Claude Mythos model autonomously identified vulnerabilities in HAWK, a candidate for post-quantum digital signature schemes, halving its effective key strength after just 60 hours of analysis. The same process took human experts roughly two years of review. The AI model didn’t stop at HAWK. Claude Mythos also accelerated attacks on a reduced variant of the AES symmetric cipher by 200 to 800 times, demonstrating that its cryptographic analysis capabilities extend well beyond a single algorithm family. Each discovery cost approximately $100,000 in compute. HAWK matters because it’s a candidate in the ongoing global effort to develop post-quantum cryptography, algorithms designed to withstand attacks from future quantum computers. Finding weaknesses in these candidates now, before they’re widely deployed, is exactly what the security community needs. Anthropic says it responsibly disclosed the findings to both the developers of the affected cryptographic methods and government partners. Now, to be clear: Claude didn’t break Bitcoin’s encryption or Ethereum’s signing scheme. HAWK is a different algorithm entirely, and the AES variant that was attacked was a reduced version,...