OpenAI’s math release draws criticism over verification and format

Mathematicians say the batch is not yet easy to trust or reuse, and the release missed requests for better links between prose and formal proofs.

OpenAI has dumped a large batch of AI-generated mathematics results, but mathematicians say the release still falls short of the standards needed for the field to trust or build on it. The company says it published nearly 400 results across more than 700 manuscripts, with 300 top-line results out of 719 manuscripts formalized in Lean, or about 42%. That leaves researchers facing a long verification job, since many claims still need human review and the formal proofs do not always line up cleanly with the accompanying writeups. The backlash also centers on process: an advisory group had asked frontier labs to avoid proprietary-model math testing and to provide machine-readable links between natural-language and formal artifacts, but OpenAI did not fully do that here. The episode matters because if these results are genuine, they could reshape active areas of mathematics; if not, they risk adding a lot of hard-to-verify noise to the literature.