The Advisory Group on Mathematics and Artificial Intelligence (AGMAI) is an independent group of nine research mathematicians, hosted at the Institute for Advanced Study (IAS) in Princeton, New Jersey, that advises AI companies on how they interact with mathematical research and with the mathematical community, including how new mathematical results are presented and released.[1][2] It was announced on September 21, 2026, in a post by OpenAI and on the group's own website, agmai.org.[1][2][3] The group says it formed after OpenAI approached some of its members about setting up an external advisory board, and that the members, in agreement with OpenAI, chose instead to create an independent body and invite others to join.[2] Its members are not paid, it says it will publish its recommendations, and it offers advice to any AI company whose models are likely to have a significant impact on mathematics.[2]

Its first assignment is specific. The group says it is "currently facing the very specific challenge of advising OpenAI on how to coordinate the release of a large number of significant results in mathematics that they report have been produced by their internal model."[2] OpenAI's announcement of the group also carried a new claim: that an internal model it began training on August 28, 2026 had "resolved more than 100 long-standing open problems across most areas of mathematics," in addition to its proposed Navier-Stokes solution.[1]

Members

OpenAI's post and agmai.org list the same nine initial members, with the affiliations below.[1][2]

Member	Affiliation as listed
François Charles	ENS-PSL
Camillo De Lellis	IAS, GSSI
Timothy Gowers	Collège de France, Cambridge
Martin Hairer	EPFL, Imperial College London
Nikhil Srivastava	Berkeley, Simons Institute
Ulrike Tillmann	Oxford, INI
Ravi Vakil	Stanford
Edward Witten	IAS
Melanie Matchett Wood	Harvard

The Verge described the members as drawn from institutions including Stanford, Harvard, Oxford, and Cambridge, with multiple Fields Medals and MacArthur fellowships among them.[7] Wood is the group's only Harvard affiliate, according to The Harvard Crimson.[6]

Formation and structure

The group's statement of purpose opens with the view that "we are at a historic moment for our discipline" and that "recent events raise urgent questions about how to support the long-term prospects for deep human understanding of mathematics."[2] It describes its role as advising AI companies "on their interactions with mathematical research and with the mathematical community, including the responsible presentation and release of mathematical results," and serving as "one possible channel of communication between mathematicians and the AI industry."[2]

The same text sets out its limits. The group "operates independently of any AI company and members do not accept payment for this work," will publish its recommendations to AI companies on its website, and has no decision-making power at any company: "the responsibility for the decisions made by any company will rest with that company."[2] The announcement was also published as a guest post on Terence Tao's blog. The group invited mathematicians to send input through a web form, saying responses would inform its recommendations and would not be made public without the respondent's approval.[2][3]

Hairer gave more detail in a September 22 post on the blog Proofs and Prompts, cross-posted to Tao's blog, which he opened by referring to "a torrent of misinformation" about the group.[4] He wrote that "the original impetus of forming this group was OpenAI's response to the statement A Severe Misalignment of AI in Mathematics," but that not all members had been contacted by OpenAI, that nobody besides the nine takes part in the group's conversations, that members receive no money, and that the IAS provides its technical support, "so far mainly regarding IT, legal, and communications." Members, he wrote, had signed no documents restricting what they say in public or whose advice they seek, "besides the obvious confidentiality requirements."[4]

In an interview with The Harvard Crimson, Wood said the members had not wanted a special relationship with OpenAI: "An independent group would be much better positioned to work for the needs of the mathematical community. We didn't want to have any special relationship with OpenAI. Whatever work we do and advice we give, we want to be able to give to any company."[6]

OpenAI's description of the mandate

OpenAI's post, titled "Advisory Group on Mathematics and Artificial Intelligence," says the group "will serve as a bridge to the mathematical community and broader public, giving mathematicians a voice in how we move forward."[1] It lists three areas of advice on emerging results: helping OpenAI assess their significance, advising on how to coordinate their dissemination, and advising on "academic and professional standards of mathematical research." The group will also advise on how OpenAI's tools can support mathematical research and learning.[1]

The post makes several commitments and one exclusion.[1]

Term	OpenAI's wording
Independence	"The group will operate independently from OpenAI."
Unrequested advice	The group may "offer advice we have not requested"
Public comment	It may "comment on OpenAI's impact on mathematics, and make its advice public"
Pay	"Its members will not be paid by OpenAI"
Membership	"the group can change its membership as it sees fit"
Exclusion	"the group will not be responsible for advising us on how to pace our internal progress on mathematics"

OpenAI called the arrangement "a first step" and wrote that "its value depends on its members being able to exercise their own judgement and challenge ours."[1] TechCrunch read the exclusion as meaning the group "won't be given leeway to slow down or redirect OpenAI's ongoing mathematical research."[5] The Verge wrote that it appeared the group would be given early access to OpenAI's research in order to do its work, and that OpenAI had not given details of what the group's stated freedoms would entail.[7]

Background

OpenAI's post places the group in a sequence of events. It says that, in addition to resolving the Navier-Stokes Millennium Prize problem, the internal model has resolved more than 100 further open problems, that "the pace of its progress in mathematics has surprised the mathematicians within OpenAI," and that this led to internal discussion of how to inform the mathematical community.[1] It then cites the open letter A Severe Misalignment of AI in Mathematics, in which "mathematicians raise concerns about the negative externalities of solving open problems as a benchmark for new AI systems," and says that "their criticisms highlight the need for thoughtful engagement of AI companies with the math community."[1]

That declaration, published on September 11, 2026 at mathandai.org, argues that "the push by AI companies to solve mathematical problems as a benchmark is detrimental to the science of mathematics, and to the mathematical community," and that such solutions are "often" announced "in a rush, leaving no time for a proper writeup, the isolation of new methods and ideas, and citing relevant previous work of others."[10] Its signatories are Fields Medallists. Hairer is the only AGMAI member among them.[10] The declaration's separate public roll of endorsers, which Hairer wrote had reached "nearly 8000 mathematicians," listed Camillo De Lellis (Institute for Advanced Study) and Ravi Vakil (Stanford University) when searched on September 23, 2026.[4][11]

The Navier-Stokes announcement of September 8, 2026 had drawn disputes over priority and credit. The Verge summarized the fallout as OpenAI "fending off accusations of scooping researchers, failing to credit the human work its models built upon, and even stealing from those using its models."[7] OpenAI's Navier-Stokes page says the effort used an internal model "significantly more capable than GPT-6 Astra" and that the model's training "is ongoing."[9]

Gowers, a member, wrote on September 17 that he had not signed the Fields Medallists' letter despite "agreeing with much of what it said," and set out his own position instead, as he had done earlier with the Leiden Declaration.[12] In the same post he disclosed that he has contacts in OpenAI's mathematics group, has been given early access to some of its models "typically only a few days before they have been released" and free access to its Pro models once released, and that he has "never been paid by OpenAI."[12]

The "more than 100 open problems" claim

The figure is OpenAI's own. Its post states that the model "has now resolved more than 100 long-standing open problems across most areas of mathematics."[1] The phrase "pace of its progress" links to a chart on OpenAI's Navier-Stokes page titled "Performance of GPT-6 Astra and our Internal Model on a curated set of open math problems," which plots pass rate against test-time compute on a log scale.[1][9] In the chart's data, GPT-6 Astra's pass rate runs from about 8% to 17% as compute increases, and the internal model's from about 26% to 47%.[9]

The post does not list the problems, give proofs, or say how "resolved" was judged, and the Navier-Stokes page does not identify the problems in the curated set.[1][9] The Decoder noted that which problems the model had solved, how it solved them, and what the results mean "remain open questions."[8] The advisory group itself describes them as results "that they report have been produced by their internal model," and treats coordinating their release as its current task.[2] According to The Verge, OpenAI had also said earlier that it was closing in on another Millennium Prize problem.[7]

Reception

TechCrunch reported the group as OpenAI's attempt to give "mathematicians more input into the company's math-oriented research," and noted that its members can offer unsolicited advice, go public with their views, and control their own membership, "giving them a measure of independence from the company."[5] The Decoder described the group as having a say "in how results are shared, not whether or how fast they're produced."[8]

The Verge spoke to researchers outside the group. Simon Machado of ETH Zurich said the arrangement seemed "shrouded in mystery" and questioned how the nine were chosen; he also said of the members, "They're amazing mathematicians, but I don't know if they are the people I want to represent me in more political questions." Francesco Fournier-Facio, an incoming professor at Heriot-Watt University, said greater engagement "can only make things better" but that "it feels like an ivory tower." Kevin Buzzard of Imperial College London questioned why OpenAI needed such a body to learn what mathematicians want: "It wasn't proofs of hard theorems, it was better understanding of our subject," he said. "It's not entirely clear to me that you need a committee of brilliant people to hammer the point home."[7]

Members addressed some of these points themselves. Hairer acknowledged that "the nine of us clearly don't form a representative sample," that AI labs "can choose to simply ignore" advisory advice, and that "it would also be very naïve to believe that the AI labs won't try to spin whatever we say in a way that suits their PR machine."[4] He gave his reasons for joining anyway. One was that recent releases of high-profile results "on privately controlled websites," with "abject levels of scholarship and only minimal presentation effort," had been "undermining the usual standards of scientific practice." Another was that refusing to engage would give OpenAI "a PR win painting the mathematical community as a bunch of out of touch luddites." He also cited the declaration's own call for the issues to be addressed "by the companies developing these technologies." He wrote that the group had not yet settled on its advice.[4] Wood told the Crimson that AI output "often includes plagiarism," "fails to present ideas clearly, fails to document the entire scientific process, and usually there's no human who is willing to take responsibility for the output," and that it was too early to say what the new norms should be.[6]

See also

References

OpenAI. "Advisory Group on Mathematics and Artificial Intelligence." September 21, 2026. https://openai.com/index/advisory-group-on-mathematics-and-ai/ (read via Internet Archive capture of September 22, 2026, 12:45 UTC: https://web.archive.org/web/20260922124518/https://openai.com/index/advisory-group-on-mathematics-and-ai/ ; announced on X at https://x.com/OpenAI/status/2102093145051943229)
Advisory Group on Mathematics and Artificial Intelligence. agmai.org (home page, including Purpose, Independence, Transparency, and Accountability, and Current Task). Read September 23, 2026. https://agmai.org/
Advisory Group on Mathematics and Artificial Intelligence. "Announcing the Advisory Group on Mathematics and Artificial Intelligence." Guest post, What's new (Terence Tao's blog), September 21, 2026. https://terrytao.wordpress.com/2026/09/21/advisory-group-on-mathematics-and-artificial-intelligence/
Martin Hairer. "Why I agreed to join AGMAI." Proofs and Prompts, September 22, 2026, cross-posted to What's new. https://terrytao.wordpress.com/2026/09/22/why-i-agreed-to-join-agmai/ (original: https://proofsandprompts.com/2026/09/22/why-i-agreed-to-join-agmai/)
Aditya Mehta. "OpenAI forms math advisory group as its AI resolves more than 100 open problems." TechCrunch, September 21, 2026. https://techcrunch.com/2026/09/21/openai-forms-math-advisory-group-as-its-ai-resolves-more-than-100-open-problems/
Francesco Palma. "Harvard Math Professor Joins Independent Group Advising AI Companies on Research Standards." The Harvard Crimson, September 22, 2026. https://www.thecrimson.com/article/2026/9/22/math-professor-ai-advisor/
Robert Hart. "OpenAI wants to consult elite mathematicians about how to not fumble again." The Verge, September 23, 2026. https://www.theverge.com/ai-artificial-intelligence/999167/openai-elite-mathematicians-panel
"OpenAI says its internal model solved over 100 long-standing math problems after just a month of training." The Decoder, September 22, 2026. https://the-decoder.com/openai-says-its-internal-model-solved-over-100-long-standing-math-problems-after-just-a-month-of-training/
OpenAI. "On the Navier-Stokes Millennium Prize Problem." September 8, 2026, updated September 10, 2026. https://openai.com/index/navier-stokes-solution/ (chart "Performance of GPT-6 Astra and our Internal Model on a curated set of open math problems" read via Internet Archive capture of September 21, 2026, 12:04 UTC: https://web.archive.org/web/20260921120424/https://openai.com/index/navier-stokes-solution/)
"A Severe Misalignment of AI in Mathematics." Math and AI, published September 11, 2026, DOI 10.5281/zenodo.22737750. Read September 23, 2026. https://mathandai.org/
"Endorsers." Math and AI. Searched September 23, 2026. https://mathandai.org/endorsers.php
Timothy Gowers. "Why I didn't sign the Fields medallists' letter." September 17, 2026, cross-posted to What's new. https://terrytao.wordpress.com/2026/09/17/why-i-didnt-sign-the-fields-medallists-letter/ (original: https://gowers.wordpress.com/2026/09/17/why-i-didnt-sign-the-fields-medallists-letter/)

Editing: Advisory Group on Mathematics and Artificial Intelligence