Mathematicians push back as OpenAI releases hundreds of AI-generated proofs
OpenAI released 722 mathematical manuscripts and proofs generated by an internal frontier model, though only about 300 were formalized. Researchers say validating the work could take years, and some reports contain nonsensical explanations, incorrect formulas, or results already known. Mathematicians warn that AI-driven releases may weaken collaborative mathematical understanding and raise concerns about cryptography and digital security.
OpenAI reportedly released 722 mathematical manuscripts, but only about 300 were formalized for computer verification. The model behind them was not named, and the prompts were withheld, despite recommendations from AGMAI. Some reports contained nonsensical explanations, invalid formulas, or already-known results. Retractions and corrections have already appeared, and OpenAI published changelogs for its GitHub repository.
Mathematicians say validating the output could take years. Because the system jumped from prompt to answer without transparent working, researchers may struggle to build on the results or gain new mathematical understanding. The episode also raises concerns about cryptography, digital security, and cryptocurrency risks.
The release could affect mathematicians, cryptographers, and technology firms by shifting scarce review labor toward checking opaque AI outputs. If errors or unverifiable claims spread, public confidence in mathematical research and digital security may weaken. Students and researchers might face pressure to adopt similar tools before norms for attribution and validation mature. The episode may also prompt clearer disclosure standards for AI-generated results, though the pace of such changes remains uncertain.