MIT held a panel discussion on November 16 titled "AI and Society: The Future of Mathematics." The four mathematicians on stage came from MIT, Harvard, Northwestern University and New York University, with science writer Kevin Hartnett moderating. The event took place in a tiered lecture hall in MIT's Building 3, and every seat was taken. MIT News published an on-site report on November 20.
A single line from Northwestern professor Bryna Kra was placed at the very top of that report:
"Math has changed more in the last two months than in the last 100 years."
She went on to quote the opening of A Tale of Two Cities, calling this the best of times and the worst of times: exciting, and frightening too.
After the "mathocalypse"
MIT News traced how the discussion came about. In the summer of 2025, a Google DeepMind model reached gold-medal level at the International Mathematical Olympiad. By 2026, models had started working on research problems with no known answer. In September, OpenAI announced that it had pointed thousands of agents at a single problem and, after 88 hours, produced a proof related to the Navier–Stokes equations, one of the seven Millennium Prize Problems. A month later, OpenAI released 722 manuscripts, claiming that AI had fully solved a batch of important open problems. The mathematics community has taken to calling the episode the "mathocalypse."
One of the panelists, Tristan Buckmaster, a professor at NYU's Courant Institute, was right in the middle of the storm. According to MIT News, he had spent a year working with collaborators at Anthropic and was close to solving the same problem when OpenAI made its announcement. Arguments followed over whether credit had been assigned properly, whether data had leaked, and how the agents had arrived at their solution.
In the same week as the panel, mathematics saw another development: Anthropic's mathematicians announced they had worked with Claude to complete a proof of the Riemann hypothesis, releasing the materials as Lean code, and a Stanford professor posted a purely human-written draft a few hours later. Both proofs are still awaiting expert review, and no conclusion has been reached.
The dispute is whether "priority" still matters
Harvard professor Michael Hopkins was blunt about the AI manuscripts. He said the papers themselves are "nonsense," full of "strange ideas," and that a gap remains between AI mathematics and human mathematics. Buckmaster added that for mathematics to carry weight, human understanding has to be part of it.
The disagreement was over how to respond. Kra argued for dropping the old rule of "who proved it first." Speed, she said, should no longer be the priority; the focus has moved away from producing proofs, and a new way is needed to assess whether a person actually understands the material. Hopkins did not fully agree: priority can go, but people still want to be pioneers, and once AI has plowed the entire field, where pioneers should head next is a hard question. Kra answered that explaining clearly how a proof came about is also a form of pioneering.
Kra helped draft this year's Leiden Declaration on Artificial Intelligence and Mathematics, signed by mathematicians in several countries, which calls for a response to the risks AI poses to mathematical research.
Journals and students feel the pressure first
The changes have already reached day-to-day work. Kra is an executive editor at a specialist journal, and she said submissions have risen sharply, most of them AI-generated or AI-assisted. The journal has hired another executive editor as a result. Some publications have begun to value the parts of a paper that explain the reasoning, and experimental journals devoted to expository articles have appeared. She also mentioned a new peer-reviewed "video journal," Mathematical Discourse, whose content consists of recordings of mathematicians giving talks.
Hiring is changing too. Kra said interviews are becoming more important: admissions, hiring and promotion can no longer rest on written work alone, and also depend on whether the candidate can explain it clearly in person. MIT assistant professor Jacopo Borga said that mathematicians used to be judged ninety percent by their journal record, and he is not sure that approach should continue.
On the student side, Hopkins said his group includes a large contingent of "AI vegetarians" who refuse to use AI, while others do use it, and he has to look after both. He mentioned a student with good intuition but weak execution who is very good at asking AI questions and has produced interesting results. In Hopkins's own words, this may be a new kind of mathematician.
The panel reached no shared conclusion. What journals should use in place of "first proof" to evaluate papers, and who should sign off on AI manuscripts and by what standard, each of the four answered differently. This was the first event in an MIT series, hosted by MITHIC and SERC with the Department of Mathematics as co-sponsor.
Sources: MIT News on-site panel report, used to verify the four panelists' remarks and the figures of 88 hours and 722 manuscripts; CocoLoop; The Decoder, used to verify what has been made public about the two Riemann hypothesis proofs.