The Simons Foundation said on Oct. 15 that it will work with OpenAI and Anthropic on a combined $15 million commitment in funding and resources to support the mathematics research community as it evaluates AI-assisted results, exchanges experience with research tools, and develops norms for how AI should enter mathematical research.
OpenAI said separately that it will provide $5 million, along with model access and compute resources.
Simons Foundation will run the program
The foundation said it will manage and distribute the shared pool of resources through open calls for proposals, workshops, and related programs. The stated goals are to help researchers assess recent AI-assisted outputs, learn and use new tools, exchange views, and establish standards for AI use that the community can follow.
Researchers supported by the foundation will also receive AI tools and computing resources from the two companies. The scope is not limited to mathematics and also covers basic science.
David Spergel, president of the Simons Foundation, said he wants the mathematics community to lead decisions on how these tools are evaluated, adopted, and used, rather than having the direction set by a single company or institution.
The $15 million total is not described as all cash
The funding figures in the announcement need to be separated carefully. The $15 million refers to the total amount of funding and resources committed through the collaboration involving the Simons Foundation, Anthropic, and OpenAI. OpenAI said its own contribution is $5 million, with additional model access and compute.
The announcement does not describe the entire $15 million as a cash donation. That means the figure should not be framed as an all-cash contribution jointly made by the two AI companies.
The focus is on methods for checking AI-assisted work
The initiative is not centered only on giving researchers access to models. Its main purpose is to build ways for the research community to examine AI-assisted work. Mathematical proofs need to be understood by experts, checked step by step, and, when needed, verified with formal tools such as Lean. A model may produce what looks like a complete derivation, but that alone does not make the conclusion reliable.
That is why the Simons Foundation listed workshops, open proposal calls, and community discussion as part of the support structure: the process is meant to include researcher review and collective judgment.
OpenAI pointed to recent math research materials released for outside review
In its own statement, OpenAI said its recent mathematics research included materials such as Lean formal proofs that can be examined externally.
The company also said that, in releasing recent work on how its models perform on math problems, it consulted a mathematics and AI advisory group from the Institute for Advanced Study. It published reasoning summaries, the number of problems attempted, and Lean formal proofs. OpenAI added that it still expects to improve the way it releases research based on feedback from the community.
Those steps show that if AI companies want their research output to be taken up by academia, showing model capability is not enough on its own. They also need to provide materials that outside researchers can inspect.
Funding does not amount to independent validation
Both the backers of the initiative and the people who may benefit from it sit within the broader AI research ecosystem. Company-provided tools and compute can expand researchers' access to advanced models, but funding by itself should not be treated as independent verification of any specific AI result.
The credibility of the program will depend on how proposals and outcomes are disclosed, whether researchers can criticize results freely, and whether comparisons can include different models as well as non-AI methods.
For academia, AI may help with exploration, organization, or the formalization of derivations, but it does not automatically replace peer review or the standards of mathematical proof. The collaboration is closer to a resource framework that gives researchers room to shape practice while tools are changing quickly. Its results will become clearer only after funded projects begin and research outputs are made public.