The Association for Human Mathematics (AHM) has asked mathematicians to stop working with OpenAI. In a statement dated October 7, 2026, and reposted as a guest post on Terence Tao's blog, the group answered OpenAI's 722-manuscript release with a blunt line: "Releasing over 700 files at once is not a demonstration of scholarship, but a demonstration of power."
That is a named-group response to the October 6 release, and it escalates a feud that has been building since September. This post covers what the statement actually says, who the AHM is, how it fits the earlier fights over Navier-Stokes claims and the Fields medalists' declaration, and what is and is not established.
TL;DR: the questions people are asking
| Question | Short answer |
|---|---|
| Who issued the statement? | The AHM Communications Working Group, reposted by Terence Tao as a guest post on October 7, 2026 |
| What is the core claim? | The release is "a demonstration of power," not scholarship |
| What is the ask? | Mathematicians should discontinue their work with OpenAI |
| Does it say the proofs are wrong? | No. It is about process, consent and norms |
| Is it a binding boycott? | No enforcement mechanism is stated; "boycott" is a media label |
| What is the trigger? | OpenAI's October 6 repository of manuscripts, openai/math |
What the AHM statement actually says
The statement is short, and every sentence is worth reading closely. It opens by noting that OpenAI is "currently defending lawsuits against accusations of illegal plagiarism, copyright infringement, and trademark dilution" and describes the October 6 release as "a repository of manuscripts purporting to contain solutions to a number of high-profile problems in mathematics."
Then come the three arguments.
Nobody asked. "Mathematicians did not ask for this work to be done." The framing is that the lab chose the targets and the timing, and the community is being handed the review burden.
It contradicts the advisory group OpenAI cites. The statement says the Advisory Group on Mathematics and Artificial Intelligence, from whom OpenAI "has claimed to derive its legitimacy," opened its advisory statement by saying frontier AI corporations should not test advanced mathematical problems on internal models. By doing exactly that, the AHM says, OpenAI "has indicated total disregard for the norms of scientific research." We covered that advisory position in our AGMAI responsible-release explainer and the earlier post on OpenAI's math advisory group and the 100-problem claim.
Volume is the message. The closing lines reject the claim that the release advances the subject and ask mathematicians to "discontinue their work with OpenAI and to return to a vision of science that centers human understanding."
Note what is absent. The statement does not name a single manuscript, does not claim any proof is incorrect, and does not set out sanctions. It is a position about how a lab should publish scientific claims and whether mathematicians should lend it their credibility.
Who is the Association for Human Mathematics?
The AHM's own site, ahmath.org, describes the group as one that "protects mathematics as a human endeavor against the threat of artificial intelligence" and works "to maintain the independence of the mathematical community from the encroachment of corporate interest." It argues that AI tools endanger undergraduate and graduate education, that journals face "an explosion in AI-coauthored submissions," and says it is organizing working groups to produce policy proposals and tools.
The site invites mathematicians who share these values to join; we could not verify any specific membership pledges, so we do not describe them here.
The AHM is distinct from two earlier documents. The June Leiden Declaration offered recommendations for mathematicians, institutions and policymakers on disclosure and responsibility. The September open letter from 25 Fields medalists, covered here, argued that racing to announce solutions harms attribution and the human transmission chain of mathematics. The AHM goes a step further by asking for non-cooperation with one named company.
How we got here: a short timeline
- May 2026: OpenAI says an internal model disproved the planar unit distance conjecture, an 80-year-old Erdős problem.
- September 8: OpenAI says an unreleased model solved the Navier-Stokes problem using roughly 10,000 agents; the community had not verified it. See our fact-check of the Millennium Prize claims.
- September 11: TechCrunch reports that 25 Fields medalists signed a letter, NYU's Tristan Buckmaster alleged pressure over crediting a collaborator, and OpenAI withdrew sponsorship of a Caltech math event after criticism.
- October 6: OpenAI publishes Sharing AI progress in mathematics and the openai/math repository.
- October 7: The AHM statement appears and is reposted by Tao.
Each step added a grievance: speed over writeup, credit disputes, and now sheer volume. The AHM statement is the first to convert grievance into a request for action.
What OpenAI released, and what it says about verification
Per our breakdown of the repo, the release contains 722 manuscripts in 372 families, an Apache 2.0 license, Lean formalizations for many results, and an explicit warning that some unformalized results could have issues. OpenAI says it is working with the advisory group to develop best practices, while the advisory group has said it does not endorse testing advanced problems on proprietary models.
Numbers circulate differently across outlets: the AHM says "over 700 files," others report 722 manuscripts, and one headline cited 377 results. The 722 and 372 figures come from OpenAI's own framing of manuscripts and families, so they are the safest to quote. For the mathematical content itself, see the expert reactions roundup and the Lean cost-collapse explainer for why formal checking changes the review burden.
Why a "power" argument matters even if the proofs hold up
The strongest counterargument to the AHM is simple: if the proofs are correct and checkable, who cares how they were released? Lean certificates reduce the need to trust the lab. That is a fair point and worth stating plainly.
The AHM's reply, implied by its text, is that mathematics is not only a list of true theorems. It is a social process of understanding, teaching, credit and review. A dump of hundreds of manuscripts shifts unpaid review labor onto the community, sets the research agenda by the lab's compute budget, and makes it hard for anyone outside the lab to decide what deserves attention. Our earlier piece on labs hoarding math solutions covered the opposite failure: results held back for strategic reasons. Together the two show a community that distrusts both secrecy and flood-style disclosure.
There is also a practical dimension. TechCrunch's September reporting describes mathematicians worrying that their use of coding agents might feed lab training data, and that labs with tens of millions of dollars of inference can race to a proof ahead of the original researchers. If that fear spreads, the labs lose collaborators and open research culture loses openness.
What this means for people building with AI
You do not need to be a mathematician to take something from this.
- Publication format is a product decision. How a lab releases claims (rumor, blog post, repo, paper) determines whether experts engage or push back. A 722-file repo invites triage, not trust.
- Verification tooling is the new currency. Lean-checked results are easier to defend. If you ship AI-generated research in any field, ship the checker or test suite with it.
- Expect consent and credit fights elsewhere. The attribution, plagiarism and data-use questions raised here parallel disputes in code, art and science writing.
- Collaboration risk is real. If your work depends on academic partners, a public call to disengage from one lab is a reputational signal worth tracking.
How to read the statement without overreading it
Two habits help when a short, forceful statement lands in the middle of a fast-moving dispute. First, separate what the text says from what outlets say about it. The AHM text asks mathematicians to discontinue their work with OpenAI; it does not announce sanctions, membership rules or a deadline, so words like "boycott" describe the effect people expect, not the document. Second, separate the process complaint from the correctness question. The AHM argues about consent, norms and scale. Whether individual manuscripts hold up is a different question that only careful review, and for many results the Lean checks, can answer.
Keeping those apart makes it easier to follow the next developments fairly, whether that is an OpenAI reply, a reviewer finding a flaw, or the advisory group clarifying its own position.
What is not known yet
- How many mathematicians actually join the AHM call, and whether any named collaborators with OpenAI step back.
- Whether OpenAI responds directly. As of this writing we did not find an official OpenAI reply to the AHM statement.
- Whether independent reviewers find errors in the unformalized manuscripts, which would sharpen the debate considerably.
- Whether the Advisory Group on Mathematics and Artificial Intelligence issues a further statement on the release.
We will update this post if any of these change. Sources for the primary text are the AHM guest post on Tao's blog and the AHM site; the OpenAI details come from OpenAI's own announcement and repository.
Related reading
- OpenAI Releases 722 Math Manuscripts: What the openai/math Repo Contains
- OpenAI math results that matter: expert reactions
- AGMAI and the responsible release of AI-generated mathematics
- 25 Fields medalists on the AI math race
- AI labs hoarding math solutions: Aaronson
- OpenAI Navier-Stokes agent swarm claim
- Will AI replace mathematicians? IEEE Big Mathematics
Details reflect public statements and reporting as of October 8, 2026 and may change as the situation develops.