Read as article
OpenAI Withdraws Three Math Preprints Over a Sign Error
By @sharedot · · 7 pages
- AI Frontier
- AI Mathematics
- Openai
Less than 24 hours after releasing 722 AI-generated math preprints, OpenAI withdrew three manuscripts for a sign error and revised 14 more.
What happened: three withdrawals in 24 hours
On October 6, OpenAI published 722 preprints in a GitHub repository describing purported progress on 372 math problems in geometry, computer science, algebra and other fields. On October 7, the company announced the withdrawal of three manuscripts because of a sign error. According to Retraction Watch, the error invalidated an argument in one manuscript and the construction used by two dependent papers. Each withdrawn paper now carries a notice explaining 'the gap,' and OpenAI says it has revised 14 other manuscripts with proof repairs, corrected statements, clearer hypotheses and dependencies, and one correction to an obsolete citation.
Why the error cascaded
Alex Townsend, an associate professor of mathematics at Cornell University, told Retraction Watch that an error cascading into three manuscripts is not surprising, and that he suspects more errors will be found. He argued OpenAI should have announced the Lean-verified manuscripts first — referring to the open-source language that checks proof steps — and issued a separate call for community help on the rest. The release mixed confirmed and unconfirmed work: OpenAI's collaborators at the Advisory Group on Mathematics and Artificial Intelligence recommended releasing without waiting for 'full formalization,' and roughly 50% of the results were released unverified, per Retraction Watch.
The evidence problem in the release itself
The original release, covered in The Rundown AI, posed about 4,000 problems to an unreleased internal model, with OpenAI putting average compute at roughly three hours of ChatGPT Pro thinking per result. A catalogue of papers had main results formalized in Lean, including the quasi-Riemann hypothesis proof — but OpenAI's scope note for that work says the formalization excludes the paper's later applications, and the company itself warns that unformalized results may contain issues. The withdrawals put that warning into practice within a single day.
A community already short on trust
Andrew Sutherland, a senior research scientist at MIT's mathematics department, told Retraction Watch he appreciated OpenAI acting quickly and called it 'the responsible thing to do,' but said it will take much more to earn back trust lost over the September 8 Navier-Stokes announcement. Retraction Watch reports more than 8,000 researchers have endorsed concerns that OpenAI rushed that announcement, and the Association for Human Mathematics urged mathematicians to discontinue work with OpenAI, calling a 700-file drop 'a demonstration of power.'
What comes next for verification at scale
Dan Roberts, a research lead at OpenAI, announced the withdrawals on X and wrote that the repo will continue to be updated with new formalizations and errata, per Retraction Watch. An OpenAI spokesperson said the errors were found during an audit, that the process is iterative, and that the company welcomes scrutiny and feedback from the mathematical community. The episode is an early stress test of AI-accelerated math: generation can scale with compute, but verification — including checking whether formalized proofs cover what a paper claims — has to grow alongside it.