Do OpenAI’s Math Claims Put Crypto Security at Risk?

WHAT HAPPENED

The latest claim of mathematical progress from a frontier AI lab has triggered a broad and heated debate on the soundness and durability of the cryptography that secures digital assets.

On Tuesday, OpenAI released 719 math results on GitHub, following September’s high-profile claimed solution of the Navier-Stokes Millennium Prize problem. With roughly 4,000 problems posed, OpenAI used an internal frontier model to attempt solutions, on average spending about three hours of ChatGPT Pro compute per result.

The results span 17 mathematical subfields, and OpenAI claimed progress on the famous longstanding open questions related to Riemann hypothesis, the Birch and Swinnerton-Dyer conjecture, and the Hodge conjecture. Out of the 719 results, none of them explicitly relates to cryptography, but there are adjacent discoveries.

One day after the math drop, Justin Drake, a senior researcher at the Ethereum Foundation who has deep expertise in cryptographic protocol and consensus design, wrote on X urging “the blockchain industry to calmy begin planning for ‘bunker mode’” as the “OpenAI drop made it clear that mathematical superintelligence is upon us.” Drake's concern is that AI will find a classical computer algorithm that can break ECDSA, the elliptic-curve signature scheme behind Bitcoin and Ethereum. Vitalik Buterin, one of the founders of Ethereum, echoed Drake's worry that the threat extends beyond “quantum-vulnerable cryptography” to “potentially AI-vulnerable cryptography.” Scott Aaronson, a theoretical computer scientist, further mentioned that, according to his unnamed sources, “AI companies have now started, gingerly and discreetly, investigating whether their latest internal models can break important cryptographic protocols and primitives.”