OpenAI claims 100+ open math problems solved, sets up a review group
Serguey Asael Shinder2 min readOpenAI published, on 21 September 2026, a note establishing an independent Advisory Group on Mathematics and Artificial Intelligence. The framing around it is the striking part. OpenAI says that since 28 August it has been training an internal model that, in addition to resolving the Navier–Stokes Millennium Prize problem, has now resolved more than 100 long-standing open problems across most areas of mathematics — a pace it says has surprised its own mathematicians.
The group is a response to friction. OpenAI points to a recent open letter, "A Severe Misalignment of AI in Mathematics", in which mathematicians raise concerns about solving open problems as a benchmark for new systems. The advisory group is to advise on the review and communication of emerging results — assessing significance, coordinating dissemination, and upholding academic standards.
Its independence is spelled out: members are not paid by OpenAI, may offer unrequested advice, may comment publicly on OpenAI's impact and make their advice public, and may change their own membership. Explicitly, the group will not advise OpenAI on how to pace its internal progress.
What it means
Treat the 100-problems claim as a claim. It is OpenAI describing its own unreleased model; there is no external verification in the note, no papers, no problem list. The most load- bearing sentence in the piece is the one asserting the result, and it rests entirely on the company's own account.
The governance move is the checkable part, and it is doing real work around a real tension. Mathematicians objecting that "solve a famous open problem" is a benchmark with externalities — for careers, for how proofs are trusted, for what a discipline is for — is the substance here. An advisory group that can speak publicly and is not paid is a genuine constraint, and the one carve-out (no say on pacing) tells you exactly where OpenAI kept control.
Watch what the group publishes, not what OpenAI announces. The value of an independent body is its dissent. If the first thing it does is release its own assessment, the arrangement is real; if it only appears in OpenAI's own posts, it is decoration.