25 Fields Medalists Warn AI Math Race Threatens Credit and Open Research
Twenty-five Fields Medalists warn that OpenAI and rival AI labs risk undermining attribution, proof verification and open mathematical research culture.
Summary
On September 11, 2026, 25 Fields Medal winners warned that AI labs racing to solve famous mathematical problems threaten attribution, verification and open research. NYU professor Tristan Buckmaster accused OpenAI of pressuring him to omit credit for an Anthropic collaborator who solved an important problem. He also questioned whether OpenAI used their Codex work to produce its own groundbreaking proof during a marathon weekend of inference. That proof remains unverified. On September 10, OpenAI withdrew sponsorship of a CalTech mathematics event after university researchers criticized the company.
The signatories said AI solutions benefit humanity only when mathematicians can understand, document, communicate and integrate them into the mathematical canon. Rushed announcements can omit full writeups, new methods and citations, raising plagiarism concerns and weakening the human chain that trains students and develops ideas. Mathematicians also fear Codex activity may feed new OpenAI models. Frontier labs can spend tens of millions of dollars on LLMs to beat original researchers to proofs, encouraging secrecy. The letter follows June’s Leiden Declaration, which recommended responses for mathematicians, institutions and policymakers, and frames the conflict as a warning for software engineering, science and creative professions.
Positives
- Twenty-five Fields Medal winners have publicly identified standards needed to preserve attribution, verification and mathematical knowledge.
- AI-generated solutions could benefit humanity when mathematicians can understand, communicate and integrate them into the mathematical canon.
- June’s Leiden Declaration provides recommendations for mathematicians, institutions and policymakers confronting LLM-generated proofs.
Risks & concerns
- Tristan Buckmaster accused OpenAI of pressuring him to omit credit for an Anthropic collaborator who solved an important mathematical problem.
- OpenAI’s groundbreaking proof, produced during a marathon weekend of inference, remains unverified.
- Mathematicians fear their Codex activity may be incorporated into new OpenAI models without adequate transparency.
- Frontier labs can spend tens of millions of dollars on LLMs to beat original researchers to proofs, incentivizing secrecy.
- Rushed AI proof announcements risk incomplete writeups, missed citations, plagiarism disputes and damage to open research culture.
- OpenAI withdrew sponsorship of a CalTech mathematics event after criticism from university researchers.