An AI model broke a post-quantum cryptography candidate in roughly 60 hours this summer, forcing its withdrawal from a federal standards competition within a day, according to an expert analysis published by The Hacker News in October 2026. The report, authored by Itan Barmes, co-founder and chief strategy officer of QIZ Security, warns that artificial intelligence is discovering cryptographic vulnerabilities faster than human researchers can, and the post-quantum schemes organizations are migrating to aren't exempt from the threat. The analysis argues that security leaders now face a different question than which algorithm to adopt: how quickly they can move again when the next breakthrough lands.
HAWK, a third-round candidate in NIST's competition for additional post-quantum signature schemes, fell to an Anthropic frontier model that uncovered a hidden symmetry in the scheme's lattice structure and converted it into an attack halving its effective security. The model time cost roughly $100,000, a price that drops every quarter, the report notes. A second lab's model found a separate attack on HAWK the same week. In September, AI coding agents set two RSA factoring records sixteen days apart, RSA-260 and RSA-896, using a classical algorithm from the 1990s ported to GPUs. That speed puts 1024-bit RSA within reach of any organization with a large GPU fleet, and such keys still exist in email signing, firmware, and appliances, according to the analysis. NIST published the first post-quantum standards in 2024, and governments have established migration deadlines between 2030 and 2035, yet many security leaders haven't started.
The report cites Scott Aaronson, a theoretical computer scientist, who wrote in October that sources he trusts indicate several AI companies have quietly begun pointing their newest internal models at cryptographic primitives and are keeping the results private. OpenAI had just published roughly 720 AI-generated math manuscripts, which Aaronson described as "one of the biggest days in the history of mathematics," with cryptography conspicuously absent from the list. In October, Justin Drake, a researcher at the Ethereum Foundation, called on the industry to prepare for "bunker mode" because AI-driven mathematics could break elliptic-curve signatures before a quantum computer does, potentially within months in the worst case. Vitalik Buterin took the opposite concern, writing that while everyone assumed curves were vulnerable and lattices were safe, there's "a good chance" lattice-based schemes, the heart of the new standards, take serious hits from the next two years of AI mathematics.
The comfortable assumption that organizations know what to leave and what to migrate to no longer holds, the report concludes. For decades, the best mathematicians failed to break the hard math problems underlying cryptography, providing a strong but unproven argument for their security. That was the only argument available. The quantum threat had a known shape with visible hardware progress, making waiting feel rational since organizations could watch qubit counts and act when the signal became clear. But that logic carried a hidden assumption: that the only threat to hard problems was a machine that doesn't exist yet. AI didn't wait. The post-quantum migration rested on a ranking, moving off RSA and elliptic curves onto lattices and hashes, but prominent Ethereum figures questioned both ends of that ranking hours apart. The target is no longer a migration with a fixed destination called post-quantum cryptography but rather the ability to move again, what NIST calls crypto agility in guidance published this year. Security teams need a continuous picture of which algorithms and keys are in use, where they exist, which hard problem each rests on, and where they can't be changed, the report states. The question isn't whether machines can do cryptanalysis anymore—they can—but who finds the next weakness first and how fast organizations can respond when it's announced on a Tuesday morning. Organizations that postponed migration planning on the grounds that quantum computers might never arrive now face a different calculus, since the models already exist and their discoveries survive expert review. The shift from project timeline to operational readiness represents a fundamental change in how enterprises must approach cryptographic risk.