Twenty-five Fields Medalists sign an open letter: the attribution, verification and research-ecosystem crisis behind AI math breakthroughs

To understand why mathematicians are this angry, look at what the letter is actually arguing over. On the surface it is about attribution and citation, but the letter spells out a deeper logic: the value in math is not just the proofs and who gets credit, but the intellectual super-structure that nourishes students, finds new questions and ideas, and integrates them into broader human civilization. That machinery runs on very concrete things — formal writeups, isolating new methods, citing prior literature, peer verification. What AI labs currently do is perfect the first step: get the result, announce it, and leave writing and verification for later. The trouble is that mathematics is a social discipline; a proof the community has not understood and transmitted is, mathematically speaking, only a candidate answer. That is what the line about AI-conceived ideas never becoming fully alive means: not a denial of AI's contribution, but a statement that without humans picking up the baton, the contribution stays suspended in mid-air.

The second layer of conflict is a flipped incentive structure, and this is where the real danger sits. TechCrunch notes the new reality: if a frontier lab sees a useful path to a discovery, it can spend tens of millions of dollars using LLMs to beat the original researchers to a proof. That pushes a group of people who used to be the most eager to share into secrecy. Both the letter and the coverage surface the same unease — some mathematicians now wonder whether their own use of Codex was in turn fed into OpenAI's new models. Notice the nature of that worry: it is not speculation about model capability but a specific concern about data flows and credit. VentureBeat, reporting on OpenAI's math result, noted the company used a swarm of roughly 10,000 agents to attack the problem, and said it could not rule out that the work benefited from a researcher's private Codex data. When the most senior people in a field start hesitating before typing their own drafts, the cost structure of open research has changed.

The third layer is institutional fallout, which has already moved past forum debate into visible consequences. First, the attribution-pressure allegation: NYU professor Tristan Buckmaster said this week that OpenAI pressured him not to credit a collaborator who works for Anthropic, and wondered whether the company had used their interactions with Codex to produce its own ground-breaking proof over a marathon weekend of inference. Second, the withdrawn sponsorship: on Thursday OpenAI pulled its sponsorship of a math event at Caltech after criticism from researchers at the university. Moves like that tend to appear once a relationship can no longer be maintained normally. Third, the letter is not an isolated event — it follows the Leiden Declaration released by a working group of mathematicians in June. According to The Next Web, that declaration was endorsed by the International Mathematical Union (IMU) and signed by Fields Medalist Peter Scholze, and it calls on mathematicians to confront how AI companies use published research without consent, bypass peer review, and threaten the integrity of proof and attribution. From a June declaration to a September open letter, the community's posture is visibly shifting from reminding to pressing.

Finally, the easiest line in the letter to skip past is also the one every industry should hear: if you do not particularly care about the cutthroat world of high-stakes mathematical proofs, do not forget that your field of interest is next. The letter states that the issues the mathematical community faces now are similar to issues other scientific and creative professions are facing, and indicate issues all of humanity might face: how to make sure that, as AI changes the way work is done, we do not lose sight of what that work was meant to achieve in the first place. Translated into product language, that is a very practical point: when AI can produce the output while you remain the one who carries responsibility, signs the work and explains the result, organizations need to close the process gap in advance — who reviews, who signs, where the line sits between source material and generated content, and how adopted AI suggestions are logged. Mathematics is simply the first field to hit this wall, because it has the strictest standards of rigor, the longest verification cycles and the clearest rules about credit. Those three conditions are converging in more and more industries at once.