OpenAI's 700 AI Math Papers Trigger Grief and Alarm Among Mathematicians

OpenAI has released more than 700 manuscripts, generated by its AI systems, that claim to resolve open problems in mathematics. The volume alone sets the release apart from a typical research paper: it lands on the field as a body of work rather than a single result, and mathematicians have spent the days since sorting through what is genuine, what is wrong, and what the episode means for how they work. A math blog collected over 100 reactions from researchers, and the range runs from interest in unfamiliar lines of attack to something closer to mourning.

The sharpest response came from Hugo Duminil-Copin, a Fields Medalist, who wrote simply, "I am paralysed." That line has circulated widely among mathematicians, in part because it captures a split reaction: some researchers describe finding new ideas in the AI output, while others describe the release as an intrusion into a discipline built on slow, human-verified proof. The blog's compilation, gathered under a title asking how much beauty has been lost, records both the technical objections and the emotional ones.

The technical stakes are concrete. Mathematics depends on proofs that other mathematicians can check, step by step, and a proof that no human has verified carries little weight regardless of who or what produced it. Seven hundred manuscripts is far more than any referee pool could absorb, so the field faces a screening problem before it faces a scientific one. If even a fraction contain valid arguments, the work of confirming them falls on people who did not ask for it.

The reaction also points to a shift in how mathematical labor gets organized. A discipline that has long rewarded deep, solitary work on a single hard problem is now being handed bulk output that may or may not overlap with what its practitioners consider interesting. Several respondents framed the loss in terms of craft and taste rather than correctness, which is harder to answer with a benchmark. The grief in the responses is not about being outperformed so much as about a way of working that may not survive the volume.

OpenAI has not said how the manuscripts were produced, whether they were checked internally, or what it intends to do with them. That silence leaves the field to adjudicate the claims on its own, and the early returns suggest few researchers are eager to take on the task. The question now is whether any of the 700 results survive independent scrutiny, and who will do that scrutiny if the answer is yes.