A day after OpenAI dumped hundreds of AI-generated manuscripts on the field, a new mathematicians’ group has issued a sharp rebuke, and urged colleagues to stop working with the company.
A group calling itself the Association for Human Mathematics (AHM) has hit back at OpenAI’s release of AI-generated mathematical manuscripts, saying the company has shown “total disregard” for the norms that keep science trustworthy, and urging mathematicians to discontinue their work with the AI lab.
The statement, which was reposted on Terence Tao’s blog as a guest post on October 7, came a day after OpenAI published a catalog of 722 manuscripts grouped into 372 result families across 16 areas of mathematics. The company says all of it was produced by an unreleased internal model.
“Mathematicians did not ask for this work to be done”
The AHM’s Communications Working Group opens by pointing out that OpenAI is currently defending lawsuits over alleged plagiarism, copyright infringement and trademark dilution, and says mathematicians never requested the work. It also takes aim at the company’s claim to legitimacy. OpenAI has said it consulted the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study when deciding how to release the results. But the AHM says that group’s own advisory statement began from the premise that frontier AI companies should not test advanced mathematical problems on internal models, and that OpenAI has ignored that central point.
The association says mathematicians have their own view of what progress means, shaped by the history and particular demands of the field, and rejects OpenAI’s assertion that the release advances the subject. It asks mathematicians and the public to treat this publication model with skepticism.
The sharpest line is about scale. Releasing more than 700 files at once, the AHM argues, is not a demonstration of scholarship but “a demonstration of power.” It closes by calling on mathematicians to stop collaborating with OpenAI and to return to a vision of science centered on human understanding.
What OpenAI actually released
The catalog includes claimed resolutions of well-known problems alongside many counterexamples to long-held beliefs. Among the headline entries is a claimed “quasi-Riemann hypothesis“: a zero-free region for the Riemann zeta function that is far wider than what was previously known, though it falls well short of the Riemann hypothesis itself. The set also includes claims on the Unique Games Conjecture and the Hodge conjecture for CM abelian varieties, and a matrix multiplication exponent that would be a big leap over the public record.
None of it has been peer reviewed. OpenAI says Lean formalizations will be added as they are completed and that some unformalized results could have issues. According to the company, the model was posed roughly 4,000 problems, and each result consumed about three hours of ChatGPT Pro compute on average, with the outputs then filtered for significance.
A community that is anything but united
The AHM’s anger sits uneasily beside the awe that greeted the release elsewhere in the field. Mathematicians reacted with shock and wonder on X: Rutgers’ Alex Kontorovich said the zero-free strip would be an instant Fields Medal if a human had done it, Cornell’s Steven Strogatz called the matrix multiplication result staggering, and Thomas Bloom, who runs erdosproblems.com, joked that mathematicians should all take a holiday before getting back to work. Even Levent Alpöge, who works with Anthropic and has clashed with OpenAI over credit, called it a hugely significant moment.
But the AHM’s concerns about motive have company. University of Connecticut’s Alvaro Lozano-Robledo asked why OpenAI would release hundreds of results at once, and described it as a hostile takeover of the mathematical landscape and a massive PR stunt. OpenAI CEO Sam Altman, for his part, struck a reflective note after the release, thanking “the machines” and generations of mathematicians.
Part of a longer-running tension
The statement lands after a string of friction points between OpenAI and parts of the math community. Last month, OpenAI shared a claimed solution to the Navier-Stokes problem from a model it said was significantly more capable than its public one. That announcement came with a dispute over credit, and OpenAI later said it hadn’t seen mathematicians Tristan Buckmaster and Levent Alpöge’s unpublished work while resolving it.
Tao himself, who reposted the AHM statement, has voiced a more nuanced position, but hinted that OpenAI didn’t behave responsibly by releasing all proofs at once. He has previously said OpenAI used only upbeat snippets of a much longer conversation with him in an advertisement, and has argued that the mathematical community should unite around its core values. He has also warned that AI-generated proofs can look flawless while hiding subtle mistakes, which makes the verification question hanging over a 722-manuscript release especially pressing.
The AHM’s argument also echoes the Leiden Declaration on AI and Mathematics, published in June and endorsed by the International Mathematical Union, which criticized corporate announcements that seek publicity on market timelines before the community’s accepted evaluation processes can take place.
What happens next
The AHM is asking mathematicians to walk away from OpenAI at a moment when the company is hiring from the field, with Fields Medalist Jacob Tsimerman taking leave to join it. Whether many mathematicians heed the call is unclear. The more immediate test is the one the AHM implicitly raises: how a field built on slow, careful verification responds to AI output that arrives faster than anyone can check it. Lean formalizations, independent write-ups of the biggest claims, and any journal acceptances will show whether the release holds up, and whether the community’s trust in the process can be rebuilt on its own terms.