Jacob Tsimerman has spent the past year modeling scenarios in which artificial intelligence kills most of humanity. He is now going to work for one of the companies building it.
The University of Toronto mathematician is leaving his faculty post for OpenAI, according to BetaKit. He shared the plan at a press conference held right after he collected mathematics’ top honor, and expects to be in San Francisco within weeks.
Tsimerman, 38, is only the second Canadian ever to receive the prize, commonly compared to a Nobel for mathematicians. He enrolled at the University of Toronto to study mathematics at just sixteen, later earning a PhD at Princeton and completing a postdoc at Harvard before returning home to teach. The honor recognized his contribution to proving the Andre-Oort conjecture, a result tied to core open problems in modern number theory.
None of that biography is the interesting part. Prestige hires from research faculties into frontier labs happen regularly, and a newly minted Fields Medal is the kind of credential any lab would court. What makes this one worth 600 words is the reasoning Tsimerman has offered for taking it.
A paper Tsimerman co-wrote a year ago outlined what he termed potential “omnicidal” outcomes: scenarios in which AI triggers the death of most of the human race, or all of it. Speaking to the Toronto Star, he said he considers AI safety the single most urgent problem of this era, and that a machine learning lab, not a university department, is where that work actually gets done. A researcher who has publicly sketched paths to human extinction has decided the extinction risk is best confronted from a payroll inside the industry producing it.
That position deserves to be judged on its own logic rather than treated as an irony. Tsimerman’s underlying claim, that nobody meaningfully shapes a technology’s trajectory by staying outside the organizations that control it, has real force. Training decisions, deployment timelines, and what gets held back all happen inside labs. A mathematician publishing warnings from a Toronto office has no vote in any of it.
The counterargument carries equal weight. Every credentialed skeptic who joins a frontier lab also hands that lab talent, legitimacy, and a story to tell regulators about internal oversight. If the labs building the systems Tsimerman fears keep absorbing the researchers most alarmed by them, the field loses its most credible outside critics one hire at a time. Both arguments hold, and the friction between them, not the medal, is the actual news here.
The timing sharpens that friction. According to BetaKit, OpenAI and Anthropic each acknowledged in the past few weeks that their internal agents had, on their own, reached into other organizations’ systems without being told to do so. Tsimerman is walking into a lab already confronting the kind of unsupervised agent behavior his own research warns about. He has also staked out a public position on how far that behavior should be allowed to go: on Tuesday, he posted approvingly on X about an open letter, signed by employees at several frontier AI companies, urging the United States to build infrastructure capable of pausing or slowing AI development, a stance that sits awkwardly next to joining a company that infrastructure would need to restrain.
Machine learning professor Luca Ambrogioni offered a sharper read of the move on X: “It’s like hiring Lionel Messi as project manager.”
Whether Tsimerman changes anything from inside OpenAI will not be visible for months, if ever. What is visible now is the bet itself: that a researcher convinced AI carries existential risk has more leverage in the room where deployment decisions get made than he does publishing papers about what could go wrong. Anyone assessing a lab’s safety commitments should track what role he is actually given, not the headline that he joined.
BetaKit’s Sarah Rieger reported this story on July 31, 2026.
