Mathematicians Cut Ties With OpenAI After Terence Tao-Led Backlash

The rift has been building for years. Terence Tao, widely considered the greatest living mathematician, began using ChatGPT to learn the Lean prover in October 2023 — a moment that suggested a harmonious blend of large language models, traditional algorithms, and human mathematicians. By April and May 2026, Tao was publicly warning that the field was generating proofs without digesting them, predicting that mathematical culture would eventually stop treating 'raw' proofs as solutions.

The September 2026 Navier–Stokes episode turned that warning into a crisis. OpenAI spent an estimated $15 Million in tokens on a result that Tristan Buckmaster — a mathematician independently near a solution — said was released prematurely and in poor quality. Buckmaster alleged that OpenAI's Sébastien Bubeck asked for a rival mathematician's name to be removed from a joint announcement, and OpenAI's own statements on whether user data influenced the model shifted over time.

The October 6 release of over 700 files, followed by the AHM's call to discontinue work with OpenAI, marks the moment the relationship moved from friction to formal severance. Whether this remains a dispute between one lab and one field, or becomes a broader template for how expert communities push back on AI vendors, is the open question the timeline leaves unresolved.