Twenty-five of the world’s most decorated mathematicians have drawn a line in the sand. These Fields Medal winners signed an open letter accusing major AI laboratories of undermining the foundational norms of their discipline. The core grievance isn’t just about speed; it’s about attribution. As companies like OpenAI and Anthropic race to solve century-old problems, they are allegedly bypassing the slow, careful work of peer review and citation that gives mathematical proofs their value.
The tension exploded recently when NYU professor Tristan Buckmaster accused OpenAI of pressuring him to omit credit for an Anthropic collaborator on a significant proof. Buckmaster also questioned whether OpenAI used its own Codex model to generate a competing solution over a single marathon weekend. The fallout was immediate. OpenAI pulled its sponsorship from a CalTech mathematics event after facing sharp criticism from university researchers. This incident highlights a growing paranoia: mathematicians now worry that sharing their working theories with AI tools might simply feed those insights back into proprietary models, allowing tech giants to beat them to publication.
The signatories argue that a proof is not just a binary right-or-wrong answer. It is a narrative that connects new ideas to the existing intellectual canon. “Often these solutions are announced in a rush, leaving no time for a proper writeup,” the letter states. Without human mathematicians to isolate new methods and cite prior work, AI-generated solutions remain dead ends. They cannot be integrated into the broader body of knowledge. The fear is that the incentive structure is shifting toward secrecy, as labs spend millions on inference compute to claim discoveries before independent researchers can verify them.
This conflict mirrors earlier concerns raised in the June Leiden Declaration, which sought guidelines for AI in mathematics. But the stakes extend far beyond academia. The letter serves as a warning for every creative and scientific field. If the transmission chain of human understanding breaks in mathematics, it will break elsewhere too. The real risk isn’t that AI will solve problems, but that it will erase the context required to understand why those solutions matter.
