OpenAI released over 370 mathematical results on Tuesday, spanning algebra, theoretical computer science, and mathematical logic. The dump follows last month's claimed solution to the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems carrying a $1 million reward. The volume and velocity are unprecedented — no human research group has ever published at this rate across this breadth of topics. The Institute for Advanced Study in Princeton, arguably the most prestigious independent mathematics institution in the world, responded with a statement that stops just short of alarm. The core concern: AI models now produce mathematical arguments that the humans prompting them cannot understand, verify, or take responsibility for. The IAS did not endorse the practice, and explicitly called for a "new paradigm that includes human understanding of mathematics as part of responsible scholarly output." The verification problem is structural, not cosmetic. Tristan Buckmaster, an NYU mathematician who was independently working on the Navier-Stokes problem, told the New York Times that mathematicians prompting AI models may be inadvertently feeding the model information that helps it reach the result — a form of contamination that makes it impossible to know what the model actually "solved" versus what it was led to. "There's likely to be a bunch of results where they take someone's work and then take it to completion," Buckmaster said. OpenAI's response was to announce an advisory collaboration with the IAS to give "mathematicians a voice in how we move forward." The company did not indicate it would pause or slow its use of frontier models on advanced mathematical problems. The advisory board has asked for "equitable access" to proprietary models for the global mathematics community — an implicit acknowledgment that the current arrangement concentrates capability in one company's hands. The two-tier risk is the sharpest concern. The IAS advisory group wrote that proprietary internal models used for mathematical research risk "effectively alienating the mathematical community from its own discipline." Mathematics has historically been one of the most open fields in science — arXiv preprints, public proofs, communal verification. OpenAI's approach inverts this: results are generated by closed models, on closed infrastructure, with closed weights, and then published as if they were ordinary research outputs. The Navier-Stokes claim illustrates the stakes. Millennium Prize Problems exist precisely because they resist solution — they are designed to be verified by the global community over years. If a proprietary model claims a solution that human mathematicians cannot independently verify or reproduce without access to the same model, the result occupies an epistemological limbo: not proven wrong, but not proven right in any way the field can trust. What OpenAI has built is a factory for mathematical claims backed by institutional reputation rather than communal verification. The 370 results may contain genuine advances. They may also contain subtle errors that no human reviewer can catch because no human fully understands the reasoning chain. The field is being asked to trust the machine and the company behind it — a request that runs directly counter to how mathematics has worked for centuries.
