OpenAI partners with independent mathematicians to vet an internal model's claim of cracking more than 100 long-standing problems.
The AMW Read
OpenAI, a segment case-study subject, stands up independent oversight in direct response to a mathematician open letter challenging its unverified benchmark claims, advancing the ongoing debate over frontier capability disclosures.
OpenAI partners with independent mathematicians to vet an internal model's claim of cracking more than 100 long-standing problems.
On September 21, OpenAI announced a partnership with the Advisory Group on Mathematics and Artificial Intelligence (AGMAI), a nine-member independent body based at the Institute for Advanced Study in Princeton, whose founding members include Fields medalists Timothy Gowers (Cambridge) and Martin Hairer (EPFL/Imperial College London). OpenAI said an internal model that began training on August 28 has, in addition to the Navier-Stokes millennium-problem claim it announced earlier in September, solved more than 100 long-standing unsolved problems across nearly every branch of mathematics, a pace the company says surprised its own in-house mathematicians.
The move follows direct pressure from the mathematics community. An open letter titled "A Severe Misalignment of AI in Mathematics," signed by Fields medalists on September 11, argued that AI companies treating unsolved problems as benchmarks harms the field by rushing announcements ahead of proper citation and attribution — concerns sharpened by the Navier-Stokes dispute, where a New York University mathematician who had been working on the same problem publicly objected to OpenAI's account, and by an October 2025 executive post claiming GPT-5 had solved an Erdos problem, later deleted after criticism. AGMAI can publish advice OpenAI did not request, pays its members nothing, and holds no approval authority — but it is now the body through which OpenAI's unverified mass problem-solving claim will be evaluated before wider publication, with member feedback currently being solicited from the broader math community.
For builders and investors, the signal is that unverified claims of AI-generated mathematical breakthroughs are becoming a reputational liability on their own; labs making comparable claims should expect pressure to attach third-party review before results are credited.
