The Simons Foundation has announced a new set of initiatives to help mathematicians navigate AI, backed by $15 million in combined funding and resources from the foundation, Anthropic and OpenAI. According to the foundation's announcement on October 15, 2026, the foundation will manage and award the money, including through an open call for proposals.
The timing is not accidental. Over the past weeks, AI-written mathematics has become one of the most contested topics in the field: OpenAI published hundreds of AI-written manuscripts, Fields medalists signed a declaration, a mathematicians' association told people to stop working with OpenAI, and a Beijing alliance launched its own AI-math effort. This programme is an attempt to give the community a funded, neutral venue to sort out what to believe and how to proceed.
TL;DR
| Question | Answer (source: Simons Foundation) |
|---|---|
| How much? | $15M in combined funding and resources from the foundation, Anthropic and OpenAI |
| Who runs it? | The Simons Foundation manages and awards the funding |
| How do researchers apply? | Through an open call for proposals (details to follow on the foundation's site) |
| What does it pay for? | Workshops and initiatives to evaluate AI-assisted results, learn tools, share perspectives, shape standards |
| Free tools? | Anthropic and OpenAI provide AI tools and compute to foundation-supported researchers, in maths and basic science |
| Not confirmed in the release | How the $15M is split between the three partners |
What the announcement says
The foundation says the funding is meant to support "workshops and initiatives that help researchers evaluate recent AI-assisted results, learn and use new tools, share perspectives and shape collective standards for AI use." Three themes stand out.
- Evaluation. The first item is evaluating recent AI-assisted results. That points directly at the backlog of claims that mathematicians have not had time or funding to check.
- Education. Researchers are to learn and use the new tools, which implies hands-on workshops rather than only position papers.
- Standards. Shaping collective standards for AI use covers questions such as disclosure, credit, formal verification and how journals treat machine-generated proofs.
Foundation President David Spergel said: "The community should be driving discussions of how these tools are evaluated, adopted and used, not any single company or institution." That sentence is a deliberate answer to critics who say labs are setting the agenda by releasing results first and asking questions later.
The role of Anthropic and OpenAI
Beyond money, Anthropic and OpenAI will "provide AI tools and computing resources to researchers supported by the Simons Foundation, extending beyond mathematics to include basic science." So the access reaches Simons-funded researchers in other fields as well, not only mathematicians. The foundation's announcement does not say which models or how much compute, and it does not give a per-lab dollar breakdown.
Any per-lab figure you may see quoted elsewhere is not in the foundation's announcement, so treat it as unconfirmed until OpenAI's, Anthropic's or the foundation's pages state it.
Why Simons, and why Lean
The foundation points to its own history: it was "an early backer of machine learning for science, including funding the development of the Lean proof assistant, which has revolutionized how mathematicians work and made AI-assisted proofs possible." Lean, an open-source proof assistant built around a minimal trusted kernel, matters here because it lets a proof be checked by a program rather than trusted on a model's say-so. Several recent AI-math stories turn on exactly that distinction between a claimed result and a Lean-checked one; see our explainer on OpenAI's math results and expert reactions and the what-to-check guide for the 722 manuscripts.
The announcement adds that the foundation has convened workshops, lectures and collaborations on AI's role in the field, and its news page links earlier pieces such as "From Trust to Verification: Lean's Impact on Mathematics" (July 22, 2026).
The backdrop: a field divided
Reading this announcement without the context would miss why it matters.
- OpenAI's math release. OpenAI published a large set of AI-written manuscripts on GitHub, which we covered in the 722-manuscript guide and in OpenAI's math results that matter.
- Pushback. The Association for Human Mathematics issued a statement criticising the release; see what the statement says.
- Fields medalists. A declaration on AI and mathematics drew attention in September: Fields medalists' AI-math declaration.
- Responsible release. A separate proposal covers how to release AI-generated mathematics responsibly: AGMAI responsible release.
- China. Shing-Tung Yau launched a Beijing alliance with 1,000+ researchers: Yau's AI Math Alliance.
Against that backdrop, a neutral funder pairing with two rival labs is a notable arrangement. Whether mathematicians who objected to working with OpenAI will accept money that is partly OpenAI's is an open question; the announcement tries to address it by stressing that the foundation, not the companies, runs the awards.
How an AI-assisted result actually gets checked
The word "evaluate" in the announcement covers several very different activities, and it helps to separate them, because the funding could go to any of them.
Formal verification. In a proof assistant such as Lean, a proof is a program, and a small trusted kernel checks every step. If the file compiles, the proof is correct relative to the stated definitions, whoever or whatever wrote it. The weak point moves to the statement: a model can prove a theorem that is not the one the humans meant, so someone still has to read the formal statement and the definitions it depends on. That review is cheap compared with checking a long informal proof line by line, which is why formalisation keeps appearing in AI-math stories.
Expert refereeing. Many results are informal manuscripts. Checking them means experts spending days or weeks of unpaid time per paper. A flood of machine-written manuscripts therefore creates a refereeing bottleneck, and workshops that train people to triage and check such work are one concrete way funding could help.
Novelty and attribution. Even a correct proof may restate something in the literature. Evaluating whether a result is new, who should be credited and how the AI's contribution is disclosed is a standards question rather than a technical one, and it is exactly the sort of thing the foundation says it wants the community to settle.
Tool literacy. The announcement also names learning and using new tools. Many working mathematicians have not yet tried a proof assistant or a coding agent on their own problems, so hands-on sessions can change what counts as ordinary practice.
None of these is solved by money alone, but each needs convening power and time, which a funder with decades in the field can supply. Readers who want the technical background on cheap formal proofs can start with our post on how Lean 4 formal proofs got cheaper.
Questions mathematicians are likely to ask
Is this a lab-controlled programme? The announcement says no: the foundation manages and awards the funding. The test will be the composition of review panels and the published selection criteria.
Does accepting the money mean endorsing a lab's results? Nothing in the announcement says so. It describes support for workshops and initiatives, and the funding goes to community proposals, but some researchers who have publicly criticised OpenAI's release may still decline to participate.
What about other labs? Only Anthropic and OpenAI are named as partners. Google DeepMind, Meta and others have also published AI-math work this autumn, for example Google's Gemini and five math problems and Meta Muse Spark's six open problems, so whether the programme stays a two-lab arrangement is a fair question.
Does the compute reach beyond maths? Yes. The announcement says tools and computing resources extend to Simons-supported researchers in basic science, so physicists, biologists and others funded by the foundation may also benefit.
What this means if you do or use mathematics
- Researchers: watch the Simons Foundation site for the open call. Proposals that evaluate specific AI-assisted results, or build shared checking practice such as Lean formalisation, fit the stated goals.
- Journal editors and conference chairs: the "collective standards" language suggests that community-drafted disclosure rules may follow. Having your venue represented in workshops is cheaper than adopting a policy after the fact.
- Builders of math tools: the funding is for the community's evaluation and learning, but free access to Anthropic and OpenAI tools for supported researchers will raise demand for workflows that produce checkable output (Lean files, reproducible computations) rather than prose proofs.
- Everyone else: treat headline claims of AI-solved problems as claims until a formal checker or independent experts confirm them; we apply that rule in posts such as the Millennium Prize fact-check.
What is not yet known
- The per-organisation split of the $15M and what counts as "resources" versus cash.
- The timeline, size and eligibility of the open call for proposals.
- Which models and how much compute researchers will receive, and under what terms.
- How conflicts of interest will be handled when labs fund evaluations of their own results.
- Whether the programme will fund formal verification work specifically.
We will update this post when the foundation publishes call details.
Related reading
- OpenAI math results that matter
- OpenAI's 722 math manuscripts: what to check
- Association for Human Mathematics statement
- Fields medalists' AI math declaration
- Yau's AI Math Alliance
- Anthropic's Claude Fermat's Last Theorem Lean proof
Details are accurate as of October 15, 2026, from the Simons Foundation announcement; the split between partners was not stated there.