Mathematics meets OpenAI's "mathocalypse": retractions and a boycott
The verification of the content has barely begun, and the disagreement over the figures in the release – 372 or 377 problems, 25 or 28 Fields Medalists – remains unresolved.
What actually happened
On October 6, 2026, OpenAI released 722 papers addressing 372 unsolved problems – from algebra and geometry to theoretical computer science. The material was largely produced by an unreleased AI model and was posted to a GitHub repository containing more than 700 files (The Conversation, Moneycontrol). The mathematics blog Proofs and Prompts has collected more than 100 responses from researchers (The Decoder).
The reactions have been swift. The release has been described as everything from a "drop" and a "dump" to "carpet bombing" and a "mathocalypse," and according to The Conversation it has sent shock waves through the field. Several papers have already been retracted or corrected.
The figures vary between sources: The Conversation reports 372 problems, while Moneycontrol writes 377. The discrepancy cannot be resolved with the available source material.
What is claimed – and what is verified?
Some of the papers claim progress toward the Riemann hypothesis and the Birch–Swinnerton-Dyer conjecture, two of the most famous unsolved problems in mathematics. But according to the mathematician Álvaro Lozano-Robledo, as quoted by The Decoder, OpenAI only managed a weaker variant of the Riemann hypothesis: "quasi-RH is huge! But it is not RH."
In September, OpenAI announced an apparent solution to one of the seven Millennium Prize Problems, the Navier–Stokes equations (The New York Times). According to the newspaper, the October release contained no new proofs of Millennium Prize Problems – only apparent progress on some of the still-unsolved ones.
The evaluation effort is, at best, in its early stages. Some papers have been retracted, others have been criticized for being hard to read, but several experts say, according to The Decoder, that they are impressed by the results. More than three dozen mathematicians The Verge has spoken with agree that it may take years to understand what OpenAI has actually delivered (The Verge).
Lean – a partial safety net
Many of the manuscripts contain formalizations in Lean, a programming language and proof system that makes it possible to verify results mechanically. But the formalizations are unevenly scattered throughout the material, and several mathematicians have complained about the lack of formalization. According to The Verge, Lean-verified results also require further checking.
Trust in Lean as a tool is not uncritical either. In the summer of 2026, the system uncovered a series of soundness bugs one after another, a period that, according to 36kr, is known in the community as the "summer of Lean soundness bugs."
The community responds – from shock to boycott
Terence Tao describes mixed feelings: the AI proofs appear to "introduce clever new ideas that will be fruitful once digested," he writes, according to The Decoder. But he is also "deeply frustrated" by the absence of humans able to discuss the work, present it, or teach students about it. According to 36kr, Tao has co-signed the statement "The Serious Misalignment of Artificial Intelligence in Mathematics," and the Human Mathematics Association, which he leads, has issued a statement urging an end to cooperation with OpenAI.
Ben Green says, according to The Decoder, that the release solved around three-quarters of the goals in a prestigious European research grant he received as recently as June. Fields Medalist Peter Scholze urges patience: "We should remember that we are all in this together; mathematics is a marathon, not a sprint; and the goal is and will always be human understanding of mathematics, which will inevitably take time" (The Decoder).
In September, 28 Fields Medalists launched a website called Math and AI with an open letter accusing AI companies of goals that are "seriously misaligned" with those of mathematicians (CNET). Curtis McMullen at Harvard writes to CNET that the rollout of mathematical work by frontier AI companies has occurred "with disregard for the deep disruption of our community and its academic ecosystem," and that the goal "appears to be to boost the valuations of these companies at all costs" (CNET). Note that 36kr gives 25 Fields Medalists for the initiative, not 28 – a discrepancy that cannot be resolved with the available sources.
The Association for Human Mathematics (AHM) has now gone further, calling on mathematicians to boycott OpenAI, on the grounds that the company is violating fundamental norms of scientific research (Moneycontrol).
OpenAI's side
OpenAI justifies the release by saying it is meant to "enable further progress in mathematics" (The Conversation). The company says it has established an advisory group at the Institute for Advanced Study, but has not explained how it will function, describing it only as "a first step" (CNET). According to CNET, OpenAI has not responded to its inquiries, and there are no details on whether the company will share its most advanced models.
What remains
The big question is how much of this actually holds up. With retractions, uneven Lean formalization, known soundness bugs from the summer, and a consensus among the mathematicians consulted that understanding may take years, no results in the October release have been confirmed as correct so far. It is also unclear how many papers have been retracted, or whether the advisory group at the Institute for Advanced Study will change how such releases happen. So far, the only certainty is that the way this work was published has created a serious split in the mathematical community.