25 Fields Medalists sign open letter slamming AI labs' rushed math benchmarks after OpenAI's Navier-Stokes claim

机器之心 · wechat · 2026-09-12

Terry Tao, Yu Deng and 25 Fields Medalists co-signed "A Severe Misalignment of AI in Mathematics," arguing that AI labs racing to solve open problems — announcing results hastily, without papers, method distillation, or proper attribution — threatens math's chain of transmission.

The trigger: On Sept 8, NYU's Tristan Buckmaster and Anthropic's Levent Alpöge announced Euler-equation blowup proofs built with Codex/Claude; hours later OpenAI claimed a full Navier-Stokes proof by an unreleased internal model (10,000 agents, 88 hours, 2.7M messages, 130B output tokens, Lean-verified, with a forcing term, prize declined). Buckmaster alleged OpenAI started only after learning of his progress and that OpenAI's Bubeck pressured him to drop Alpöge's name; OpenAI disputes the timeline.

Context: OpenAI's Erdős distance-counterexample in May, the Leiden Declaration in June, Alpöge's Jacobian conjecture result in July, and Fields winner Tsimerman joining OpenAI. Chollet warns math students increasingly say "I don't want to be a mathematician anymore," and some suspect their Codex logs fed models — pushing open math culture toward secrecy. The letter has no enforcement power but offers a reference point for journals, funders, and policymakers.

Related event: Terence Tao and 25 Fields Medallists Warn of Severe AI-Mathematics Misalignment(37 posts)→

Original post →

More from AGI Musings

AGI Musings channel →