OpenAI Forms Advisory Group on Mathematics and Artificial Intelligence after AI Resolves More Than 100 Long

OpenAI Forms Advisory Group on Mathematics and Artificial Intelligence after AI Resolves More Than 100 Long-Standing Open Problems

OpenAI claims that a new internal model has resolved more than 100 long-standing open problems across most areas of mathematics.

The same system is the one OpenAI earlier tied to a proposed solution of the Navier–Stokes Millennium Prize problem. The company said the pace surprised its own mathematicians. Most of the 100-plus results have not been released for outside review.

To handle that backlog, OpenAI is working with an independent Advisory Group on Mathematics and Artificial Intelligence, hosted at the Institute for Advanced Study in Princeton. The group will advise on whether results matter, how to publish them, and how to keep academic standards. It will also advise on tools for research and teaching. It will not tell OpenAI how fast to run internal math work. Members are unpaid. They can give unsolicited advice, criticise the company in public, and control their own membership. They have no vote over OpenAI’s product timeline.

Who Sits on the OpenAI Mathematics Advisory Group?

The first nine members are François Charles (ENS-PSL), Camillo De Lellis (IAS and GSSI), Timothy Gowers (Collège de France and Cambridge), Martin Hairer (EPFL and Imperial College London), Nikhil Srivastava (Berkeley and the Simons Institute), Ulrike Tillmann (Oxford and the Isaac Newton Institute), Ravi Vakil (Stanford), Edward Witten (IAS) and Melanie Matchett Wood (Harvard). Gowers and Hairer are Fields Medalists. Witten is a central figure in mathematical physics.

Some accounts say mathematicians declined a tighter in-house board, formed an independent group, then took OpenAI as a first client. The Institute for Advanced Study has said the panel can advise companies but does not hold decision-making power at any AI lab. Responsibility for what OpenAI ships stays with OpenAI.

Why Did OpenAI Set up the Droup Now?

The move follows a fight over how labs claim famous theorems. On 8 September OpenAI said a multi-agent run of about 10,000 concurrent systems produced a Navier–Stokes existence-and-smoothness result in roughly 88 hours, then spent more hours formalising it. Fields Medalists and other researchers had warned labs against planting flags on Millennium problems before the community could check proofs. OpenAI had already held a closed meeting with mathematicians in August. The new group is a named channel for review and release, not a claim that the field has signed every proof.

Verification is the other half of the story. Finding a candidate proof can now be faster than reading one. OpenAI has published selected earlier results from an internal Astra-class model, including work on Erdős-type problems. The September claim is larger and mostly still inside the company. Until papers, Lean formalisations or referee reports appear, “resolved” is OpenAI’s word, not a journal’s.

What the Advisory Group Will Do Next?

The first practical task is a queue: which of the 100-plus results to release, in what order, and with what human write-up so specialists can look for gaps. OpenAI says it wants mathematicians to use the tools on questions they already care about. The limit written into the charter matters as much as the famous names. The panel can slow a press cycle. It cannot slow the training run.

The durable facts are narrow. OpenAI began training a new internal model on 28 August 2026. It says that model has now closed more than 100 open problems and produced a Navier–Stokes claim. Nine senior mathematicians at IAS will advise on how those claims meet the field. The proofs still have to survive other mathematicians, not only an advisory letterhead.

Ahmed Banafa is an expert in new tech with appearances on ABC, NBC , CBS, FOX TV and radio stations. He served as a professor, academic advisor and coordinator at well-known American universities and colleges. His researches are featured on Forbes, MIT Technology Review, ComputerWorld and Techonomy. He published over 100 articles about the internet of things, blockchain, artificial intelligence, cloud computing and big data. His research papers are used in many patents, numerous thesis and conferences. He is also a guest speaker at international technology conferences. He is the recipient of several awards, including Distinguished Tenured Staff Award, Instructor of the year and Certificate of Honor from the City and County of San Francisco. Ahmed studied cyber security at Harvard University. He is the author of the book: Secure and Smart Internet of Things Using Blockchain and AI.