AI security model finds fatal flaw in quantum-resistant algorithm vying for US standard

An algorithm designed to protect against quantum computer attacks has been eliminated from contention as an official US cryptographic standard after an AI-powered security tool identified a critical flaw. The algorithm, known as HAWK, is a digital signature scheme that had advanced through two rounds of rigorous testing by the National Institute of Standards and Technology (NIST) in its effort to standardize post-quantum cryptographic techniques. The third round of evaluation—meant to catch precisely this sort of vulnerability—succeeded with assistance from Mythos, an AI security model developed by Anthropic.

The Discovery and Withdrawal#

Anthropic announced its findings on Monday, and by Tuesday, HAWK's developer had formally withdrawn the algorithm from the standardization process. The Mythos model identified mathematical weaknesses in the algorithm's underlying structure that rendered it broken for its intended purpose. This marks a concrete validation of AI-assisted security research in cryptanalysis.

What Mythos Actually Accomplished#

Anthropic publicized its use of Mythos to uncover weaknesses not only in HAWK but also in the widely deployed AES cipher. However, the results warrant important context. The improvements discovered are incremental rather than revolutionary—they reduce the computational effort needed to compromise these systems but don't render them immediately insecure in real-world deployment. Additionally, the cryptosystems tested were intentionally weakened versions used as part of formal peer review, not the full production-strength specifications. The underlying mathematical foundations that support these cryptosystems remain resistant to attack, at least for now.

Why This Matters#

While the findings don't represent an immediate threat to systems currently protecting sensitive data, they illustrate emerging capabilities in using machine learning to identify subtle mathematical vulnerabilities. As quantum computing advances, the ability to efficiently identify weaknesses in next-generation cryptographic methods becomes increasingly valuable. The HAWK withdrawal demonstrates that NIST's evaluation process—now enhanced by AI-assisted analysis—is functioning as intended to screen out flawed candidates before they become standardized.