Mathematicians Grapple With OpenAI’s Mass Release of AI-Generated Proofs

OpenAI released more than 700 AI-generated manuscripts on October 6, 2026, claiming solutions to hundreds of open math problems. The math blog Proofs and Prompts then gathered more than 100 researcher responses, showing a field trying to process both the possible breakthroughs and the way they arrived.

The release follows earlier AI math milestones cited by The Decoder, including the Unit Distance Conjecture, Navier-Stokes, and ten solutions presented with OpenAI’s Astra model. The immediate takeaway: the discussion is no longer just whether AI can help with math, but how the field should absorb AI-generated work at scale.

Several mathematicians said the results may contain valuable ideas, but the substance is still too early to fully assess. Some papers have already been withdrawn, and others were criticized as difficult to read.

Peter Scholze urged patience and emphasized that mathematics remains a long process aimed at human understanding. Terence Tao described the proofs as potentially containing clever new ideas, while also expressing frustration that no human authors were available to explain, teach, or discuss the work.

Much of the response focused less on individual theorem claims and more on what the release means for mathematicians’ lives and research identities. Hugo Duminil-Copin wrote that he was “paralysed,” while others described anger, grief, anxiety, and a loss of purpose.

Early-career researchers appear especially exposed: one PhD student said OpenAI claimed to have proved the problem at the center of his dissertation, forcing him to revise research plans while preparing postdoc applications. The clear concern is that AI labs can disrupt years of work faster than the community can verify, contextualize, or teach the results.

Researchers are divided over what remains distinctly human in mathematics. Some see a “mathematical wonderland” where AI accelerates discovery; others fear a future where mathematicians become interpreters of machine output rather than originators of research agendas.

The article notes that tensions were already rising: the Association for Human Mathematics had called for an OpenAI boycott, and 25 Fields Medalists had warned about misalignment between AI industry goals and mathematics. The key question now is not only whether AI can solve problems, but whether the mathematical community can keep control over standards, credit, and purpose.