The Proof Problem: Why the AI-Math War is a Warning for All Knowledge Work

The current friction between frontier AI labs and the mathematical community is being framed as more than a dispute over credit; it is a fundamental clash over the nature of discovery. In my view, this is the opening salvo in a larger battle over whether synthetic intelligence can truly reason or if it is simply perfecting the art of the plausible lie.

As TechCrunch first reported, twenty-five Fields Medalists—the most prestigious prize winners in the field—have signed an open letter warning that AI labs are endangering intellectual work in a race to solve famous mathematical problems. The signatories argue that for AI-driven solutions to benefit humanity, they must be communicable and understandable to the math community. They caution that solutions are often announced in haste, bypassing the necessary steps of citing previous work, isolating new ideas, and providing proper write-ups. Consequently, a recent proof produced by OpenAI remains unverified.

This tension has already manifested in specific institutional conflicts. TechCrunch reports that OpenAI withdrew its sponsorship of a CalTech event on Thursday following criticism from researchers at the university. Furthermore, NYU professor Tristan Buckmaster has accused OpenAI of pressuring him to omit credit for a collaborator from Anthropic who helped solve a significant math problem. Buckmaster also questioned if OpenAI utilized work performed with Codex to generate a groundbreaking proof during a concentrated weekend of inference.

Beyond individual disputes, there is a systemic fear regarding the future of open research. TechCrunch notes that mathematicians are becoming paranoid that their use of Codex is being fed back into OpenAI's models. Because frontier labs can invest tens of millions of dollars to use LLMs to beat human researchers to a proof, there is a growing incentive for secrecy over collaboration.

This crisis follows the June release of the Leiden Declaration, a document from a working group of mathematicians that provides recommendations for policymakers, institutions, and the math community to navigate the shift toward LLM-generated proofs. The core of the argument is that the value of mathematics lies not just in the final proof, but in the "intellectual super-structure" that supports students and integrates new ideas into civilization.

As the signatories of the open letter warn, this is not a niche problem for mathematicians. They argue that the issues facing their field are mirrored in other scientific and creative professions, signaling a broader human challenge: ensuring that as AI transforms the process of work, the original purpose of that work is not lost.