The Association for Human Mathematics’ Communications Working Group has urged mathematicians to stop working with OpenAI after the company released hundreds of AI-generated mathematical manuscripts.
OpenAI’s October 6, 2026 release comprised 722 manuscripts spanning 372 groups of results, according to The Indian Express. The newspaper reported that OpenAI said the work, produced by an internal frontier model, included solutions to or significant progress on hundreds of open problems in mathematics and theoretical computer science.
AHM objects to how the work was produced and released, particularly the use of proprietary models inaccessible to the wider scientific community. The statement marks an organized backlash, though it does not establish broad support among mathematicians for cutting ties with OpenAI. It also makes no determination about whether any particular manuscript is correct or incorrect.
AHM Calls for Mathematicians to Discontinue Work With OpenAI
The statement closes with a direct demand:
“We urge mathematicians to discontinue their work with OpenAI and to return to a vision of science that centers human understanding.”
The Communications Working Group is asking mathematicians to end their work with the company. Its demand goes beyond corrections, additional documentation, or a slower publication schedule. The statement does not specify which kinds of work it means, describe an implementation process, or identify researchers who have already acted on the call.
Its argument begins with an objection to the research itself: “Mathematicians did not ask for this work to be done.” That is the working group’s characterization, not a finding from a survey of the mathematical community.
The group argues that OpenAI ignored a central objection raised by the separate Advisory Group on Mathematics and Artificial Intelligence, or AGMAI. AHM describes that decision as showing “total disregard for the norms of scientific research,” which it says keep mathematics trustworthy, ethically researched, and in the public interest.
It also disputes the company’s framing of the release:
“We reject OpenAI's assertion that this release advances our subject, and we urge mathematicians and the public to view the value of this publication model with due skepticism.”
AHM’s criticism extends to the model of scientific progress behind the publication. In the group’s view, producing mathematical output at scale is insufficient if the process does not center human understanding and the field’s research norms.
The signatory was the Association for Human Mathematics Communications Working Group. Its position should not be presented as a declaration by the entire mathematics profession.
The Advisory Group Opposed Proprietary-Model Research
AGMAI’s September 29 recommendations, titled Responsible Release of AI-Generated Mathematics, support AHM’s reference to earlier advice.
The advisory group acknowledged that some frontier AI laboratories were already testing advanced mathematical problems on proprietary models unavailable to the broader scientific community. It stated its objection directly:
“We want to state clearly from the start: we do not endorse this practice, and we ask them to stop testing advanced mathematical problems on proprietary models.”
The objection is specific to advanced mathematical research conducted with proprietary models that the wider community cannot access. It is not opposition to all uses of AI in mathematics.
According to The Indian Express’s account of OpenAI’s announcement, the company said it had been consulting an independent advisory group to develop best practices for sharing AI-generated mathematical research. AHM argues that invoking that consultation does not resolve the advisory group’s underlying objection to how the research was conducted.
AGMAI’s document takes two positions: it asks laboratories to stop proprietary-model testing, and it offers recommendations for responsibly releasing mathematical output they nevertheless produce. Those recommendations are conditional guidance, not an endorsement of the activity.
The guidance distinguishes between papers that a responsible mathematician fully understands and output that the people prompting an AI system do not yet understand. For the first category, AGMAI recommends traditional mathematical publication practices: posting a preprint, submitting to journal peer review, and explaining the work in talks.
For the second, it calls for literature searches and attribution, readable exposition, documented revisions, and disclosure of information about the model and the process used to produce the results. It also recommends formalizing proofs where possible and clearly stating their formalization status where that has not happened.
Large releases carry further responsibilities. AGMAI asks laboratories to explain how problems were selected and how many comparable problems their models attempted but failed to solve. Selected successes alone cannot establish the reliability or breadth of a model’s mathematical performance.
The advisory group also says laboratories should support the development of human understanding, including through funding. Accepting such support, it specifies, would not confer legitimacy on the laboratories’ practices.
AHM and AGMAI are distinct organizations with distinct positions. AHM’s Communications Working Group called for mathematicians to discontinue work with OpenAI. AGMAI’s earlier document opposed proprietary-model testing while setting out responsibilities for releasing and understanding the resulting output.
The Scale of the Release Does Not Settle Its Mathematics
AHM makes the volume of the release part of its criticism:
“Releasing over 700 files at once is not a demonstration of scholarship, but a demonstration of power.”
That is an interpretation of the publication strategy, not a technical assessment of the proofs.
The reported 722 manuscripts spanning 372 groups of results should not be read as 722 independently verified solutions to separate open problems. OpenAI’s claim, as reported by The Indian Express, encompasses both solutions and significant progress, which are different levels of claimed achievement.
Neither the file count nor AHM’s criticism establishes the mathematical status of an individual result. The statement offers no manuscript-by-manuscript evaluation, identifies no particular incorrect proof, and does not demonstrate that the collection is valid.
Behind the criticism is a substantive research-policy question. AGMAI’s recommendations emphasize that mathematical authors should understand their arguments, verify them, and take responsibility for their content. They also recognize that AI can produce arguments that the person prompting the system cannot understand or verify.
Under that framework, publishing documents and developing shared mathematical understanding are separate responsibilities. A repository makes arguments available for inspection; that act alone does not establish that researchers understand or accept them.
Sources
- urged mathematicians to stop working with OpenAIahmath.org
- The Indian Expressindianexpress.com
- AGMAI’s September 29 recommendationsagmai.org
- republished the statement on October 7terrytao.wordpress.com
