OpenAI has withdrawn three of the 722 research preprints it published on GitHub on October 6, Retraction Watch reported on October 8, after a sign error invalidated a central argument and, with it, two results that depended on that construction. The withdrawal came less than 24 hours after the mass release, which OpenAI had billed as an effort "to push the frontier of human knowledge and enable further progress in mathematics."

The preprints described purported progress by an internal AI model on 372 unsolved problems across geometry, computer science, algebra, and other fields. The move drew immediate scrutiny from research communities — and, as we covered earlier this week in our report on the initial release of the 722 math manuscripts, the company framed the drop as an unprecedented demonstration of what its unreleased frontier model can produce. Follow the latest on this story in our latest AI developments coverage.

A Sign Error Cascades Through Dependent Proofs

On October 7, OpenAI announced the withdrawal of three manuscripts because of a sign error. According to the announcement, 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 Each withdrawn paper now carries a notice explaining "the gap."

The company also said it has revised 14 other manuscripts with proof repairs, corrected statements, clearer hypotheses and dependencies, and one correction to an obsolete citation. Dan Roberts, a research lead at OpenAI, announced the withdrawals on X and wrote that the company will "continue to update the repo with new formalizations and with any errata we notice."

An OpenAI spokesperson told Retraction Watch the company identified the errors during an audit, describing the process as iterative "as with other research manuscripts." The spokesperson added that OpenAI welcomes scrutiny and will "work to correct [errors] promptly and withdraw papers if no fixes can be found."

Notably, an OpenAI spokesperson said the company's collaborators at the Advisory Group on Mathematics and Artificial Intelligence (AGMAI) recommended releasing the results without waiting for "full formalization" — and that roughly 50% of the results were released unconfirmed.

Mathematicians: Quick Fixes, Lost Trust

Reactions from research mathematicians were measured but pointed. Alex Townsend, an associate professor of mathematics at Cornell University, told Retraction Watch that an error cascading into three manuscripts was not surprising, and that he suspects more errors will be found. "Given the skepticism of AI in the mathematics community, I think that OpenAI should have announced the manuscripts that were Lean verified first," he said, referring to the open-source programming language that mechanically checks math proofs. A separate announcement could then have invited community help verifying the rest, he argued.

Andrew Sutherland, a senior research scientist in MIT's mathematics department, said he appreciated OpenAI acting quickly after finding the errors, calling it "the responsible thing to do." But he emphasized that many mathematicians remain "unhappy with how OpenAI has behaved up to this point," pointing in particular to the company's September 8 announcement that an AI system had solved the Navier-Stokes problem — a claim a group of leading researchers criticized as rushed, "leaving no time for a proper writeup, the isolation of new methods and ideas, and citing relevant previous work of others." More than 8,000 researchers have endorsed those concerns.

The Association for Human Mathematics went further in a statement released October 7: "Mathematicians did not ask for this work to be done. Releasing over 700 files at once is not a demonstration of scholarship, but a demonstration of power. We urge mathematicians to discontinue their work with OpenAI and to return to a vision of science that centers human understanding."

What This Means for AI-Generated Science

The episode is shaping up as an early test case for how AI labs should handle machine-generated research at scale. OpenAI's approach — release fast, audit in public, withdraw what fails — contrasts with the formal-verification-first approach many mathematicians prefer, in which results are checked in a proof assistant like Lean before any announcement.

There are reasonable arguments on both sides. Waiting for full formalization could delay useful results by months, and OpenAI says it released with AGMAI's blessing. But the cascade effect — one sign error taking down two dependent papers — illustrates why verification before publication has been the norm in mathematics, and why more errors are likely to surface as the community works through the remaining roughly 700 manuscripts.

For now, OpenAI says it wants to learn from the feedback and will follow AGMAI's protocol "as best we can" going forward. Whether that is enough to rebuild credibility with the mathematical community is another question. As Sutherland put it: "I think the withdrawals and corrections will be viewed positively by the mathematics community, but it will take a lot more than that to earn back the trust they have lost."

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →