Fields Medalist in Tears After OpenAI Publishes 700+ AI-Written Math Manuscripts

Hugo Duminil-Copin, winner of the 2022 Fields Medal, has publicly admitted that he broke down in tears over news from OpenAI. The company has released more than 700 math manuscripts written by AI. According to Jiqizhixin, 16 problems in this collection lie in Duminil-Copin's own field. "They completely destroyed my field of research," the mathematician said. He added: "On Tuesday evening I cried and woke my daughter."

The line sounds like an emotional breakdown, but it points to a very concrete question: what is left for humans in theoretical mathematics if a model can close problems that took them years to work on?

What is known so far

The source material breaks off mid-sentence in its passage about the 16 problems, so some details remain unclear. We do not know how many of the 700+ manuscripts have been checked by specialists, or how many contain complete proofs. It is also unknown how OpenAI selected the texts for publication and in what format they were released. The only things on the public record are the collection itself, the number of problems that touch Duminil-Copin's field, and his reaction.

The key word in the report is "manuscripts." These are not peer-reviewed journal articles but texts whose correctness has yet to be assessed.

Why this particular person's reaction carries so much weight

Duminil-Copin received the Fields Medal for his work on the probabilistic theory of phase transitions in statistical physics. His results on the Ising model and percolation are considered classics of modern mathematical physics. This is a field where intuition and delicate estimates long seemed to be human territory.

When a specialist of this caliber talks about the "destruction" of a field, he most likely means not individual problems but speed. A series of open problems that the community has worked on for years could be closed as a batch in a single release.

Context: from olympiads to open problems

The road to this point was short. In 2024, Google DeepMind's AlphaProof reached silver-medal level at the International Mathematical Olympiad. In 2025, experimental models from OpenAI and DeepMind achieved gold-level results. The next step was open problems rather than school-level ones.

Caution is warranted here. In autumn 2025, OpenAI representatives said that GPT-5 had "solved" a number of Paul Erdős's problems. It then emerged that the model had found existing solutions in the literature, and the claim had to be walked back. The lesson for the current story is simple: between "AI produced a text" and "AI proved a new theorem" lies verification.

Key facts

• Source of the reaction: Hugo Duminil-Copin, 2022 Fields Medal winner, France.
• Source of the news: an OpenAI publication containing a collection of AI-written manuscripts.
• Volume: more than 700 math manuscripts.
• Overlap with his field: 16 problems.
• Verification status: the source does not mention independent verification.
• Publication date of the material: October 12, 2026.

How to read claims like this

Look for formal verification. If the manuscripts come with proofs in Lean or another verifier, that is a strong signal. If they do not, you are looking at a hypothesis, not a result.

Check for novelty. You need to find out whether the model is reproducing already known solutions, as happened with the Erdős problems. That will take specialist mathematicians, not a press release.

Watch how colleagues in neighboring subfields respond. A single emotional quote is no substitute for the community's assessment. If several groups confirm the correctness of even some of the 16 problems, the story will have to be taken seriously. If few confirmations emerge, this will be yet another case of an overly loud claim.

For those working in mathematics and adjacent sciences, it makes sense to start learning formal verification tools now and to get used to treating AI as a co-author whose conclusions must be checked as rigorously as anyone else's paper.