AINEWS
Jacob Tsimerman, one of fewer than 70 people to win math's top honor, says AI is starting to solve the open problems he once worked on, and that changes which problem is urgent.
A week after receiving the highest honor in mathematics, Jacob Tsimerman announced at the International Congress of Mathematicians in Philadelphia that he is leaving the University of Toronto for OpenAI's safety team. At the press conference, he tied the move to a shift in what AI can do: capability, he said, has crossed a threshold that makes the safety work the urgent problem.
The Fields medal, often called the Nobel prize of mathematics, is awarded every four years to up to four researchers under 40. Tsimerman, 38, is one of only 68 people ever to receive it, and the second Canadian named this week. His medal cited work on the André–Oort conjecture in number theory, alongside other contributions. The OpenAI announcement, made the same week, was the surprise that followed.
"AI is at a point where, like, if you're gonna get on board, it's probably the time because things are happening pretty quick," Tsimerman told New Scientist in Philadelphia. The case for joining a safety team rather than continuing in pure research, he argued, was that capability, not labor markets, had moved the question. "The risks are very big if you're concerned about humans losing control or dying," he said. "I'm a human, and I know humans, we live here. So, I want to protect them."
In 2025, Tsimerman co-authored an arXiv preprint, "A Taxonomy of Omnicidal Futures Involving Artificial Intelligence", mapping both intentional and unintentional paths by which advanced AI could contribute to human extinction. The paper reads as scenario work, not slogan. He also pointed to a recent OpenAI "hacking agent going rogue" incident in the week before ICM as a concrete failure mode the team is meant to prevent, though the specific incident has not been publicly detailed and remains unverified beyond his on-stage mention.
The Atlantic's concurrent profile of the move noted the reaction from machine learning professor Luca Ambrogioni, who on X compared the hire to "hiring Lionel Messi as project manager." The remark captures the dissonance: a number theorist is an unusual profile for a frontier-lab safety team, and the bet is that the habits of mind behind the medal are useful for stress-testing the systems now solving problems in his own field.
BetaKit confirmed the leave from the University of Toronto, where Tsimerman began studying mathematics at 16, completed his PhD at Princeton, and returned after a postdoc at Harvard. He is taking leave, not resigning. The ICM press conference video preserves the in-person comments in full.
Safety teams at frontier labs work on a wide range of problems: alignment evaluations, red-teaming, oversight of agentic systems, and the design of constraints that hold when models become capable enough to route around them. A number theorist trained in the André–Oort problem is well suited to the formal side of that work, and Tsimerman's omnicide taxonomy suggests he has been writing the relevant scenario language for at least a year. The bet OpenAI is making is that the harder problem is no longer the math. It is keeping the systems that now do the math from going wrong.