OpenAI triggers backlash from mathematicians with massive dump of AI proofs

OpenAI's massive results drop

OpenAI has sparked a fierce backlash from the international mathematics community after dumping more than 700 files online containing AI-generated proofs for hundreds of unsolved mathematical problems.

The release, produced largely by an unreleased artificial intelligence model, covers 372 open mathematical problems. It includes claimed advances on legendary, high-profile challenges like the Riemann hypothesis and the Birch and Swinnerton-Dyer conjecture—both of which carry a $1 million prize for a solution.

But rather than celebrating, mathematicians have reacted with anger and deep skepticism. A group called the Association for Human Mathematics (AHM) issued a blistering statement, which was quickly endorsed by the prominent Fields Medal-winning mathematician Terence Tao. The group rejected OpenAI’s claims of progress, calling the sudden drop of over 700 files "a demonstration of power" rather than a contribution to scholarship.

Unintelligible papers and 'hallucinations'

The release has been described by researchers as a "mathocalypse" and a "carpet bombing". The immediate issue for academics is that many of the proofs are virtually unintelligible to human readers. One senior researcher told Melissa Lee, writing for The Conversation, that a paper on his favourite problem was so incomprehensible that any journal editor would have thrown it straight in the bin.

Even OpenAI's own model, GPT-5.6 Sol, reportedly described one of its high-profile results as "a serious hallucination" when prompted to audit it. This leaves human mathematicians with the arduous task of clean-up and verification. While OpenAI claims to have "formalised" 300 of the results using the Lean programming language to allow computer verification, the process of automatic translation is notoriously prone to errors. Indeed, OpenAI has already retracted three of the papers due to basic mistakes and had to amend several others.

Disregard for scientific norms

The dispute goes deeper than poor writing. OpenAI claimed it consulted with its independent advisory body, the Advisory Group on Mathematics and Artificial Intelligence. However, that group had specifically urged the tech firm to stop testing advanced mathematical problems on proprietary, closed-source models. The AHM accused OpenAI of ignoring this central advice while using the group's name to claw back legitimacy.

This is not OpenAI's first mathematical controversy. In September, the company claimed a solution to a case of the Navier-Stokes problem, leading to accusations from mathematicians Tristan Buckmaster and Levent Alpöge that OpenAI had accessed their research data and spent $15 million in computing power to scoop them. OpenAI denied those allegations.

Observers suggest OpenAI's rush to publish math proofs is less about advancing science and more about proving its progress toward artificial general intelligence (AGI) ahead of a highly anticipated public valuation. But for the researchers left to sift through the digital pile of unvetted papers, the stunt has only damaged trust.