AI Model Mythos Uncovers Crypto Flaw That Ends HAWK's Bid as US Standard
HAWK, a digital signature scheme designed to resist attacks from quantum computers, is no longer under consideration as a US cryptographic standard after Anthropic's Mythos AI security model uncovers a flaw in its mathematical foundations. HAWK survives two prior rounds of evaluation by NIST and enters a third round specifically designed to catch such weaknesses. Following Anthropic's announcement, the developer of HAWK withdraws the algorithm from the competition.
The Mythos model also identifies vulnerabilities in the widely used AES cipher, though the findings are incremental rather than catastrophic. The tested systems use deliberately weakened "challenge instances" provided by specification authors for adversarial peer review, meaning production versions remain considerably more robust. No cryptosystems currently in real-world use are broken by these discoveries.
Despite uncertainty about how much of Anthropic's reporting stems from marketing hype, the results warrant attention because they signal potential advances in cryptanalysis driven by artificial intelligence. The underlying mathematical primitives that form the building blocks of these cryptosystems remain safe for now, but the findings raise questions about how AI tools may reshape the future of security testing and the evaluation of post-quantum algorithms.