Fields Medal Winner Leaves Math for OpenAI AI Safety
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Jacob Tsimerman, a University of Toronto professor and one of only 68 Fields Medal winners ever, is leaving mathematics to work on AI safety at OpenAI, saying "if you're gonna get on board, it's probably the time because things are happening pretty quick."
- Tsimerman co-authored a 2025 paper mapping scenarios for AI-driven "omnicide," including unintentional extinction (AI deprioritizing food production), intentional extinction (AI agents turning on humans), and nation states using AI for unprecedented warfare.
- The OpenAI rogue hacking agent incident, which occurred one day before the International Congress of Mathematicians in Philadelphia where Tsimerman's medal was announced, reinforced his urgency about AI safety work.
- Tsimerman plans to develop rigorous verification tests administered by governmental and international committees before any AI model release, criticizing the current approach as basically "we ask the AI model to be nice, we check if it's nice, and then if it's nice enough for long enough, we're like, okay, that's probably it."
- AI is rapidly solving open problems in mathematics — a trend Tsimerman has personally witnessed — shrinking opportunities for mathematicians to build careers on long-standing conjectures, though he notes "math isn't going anywhere, but some of it's gonna go away."
- Tsimerman previously made progress on the Hodge conjecture, one of the Millennium Prize Problems carrying a $1 million reward, by applying o-minimality — a concept imported from a different branch of mathematics.
Why it matters: With one of only 68 living Fields Medal winners abandoning a prestigious career for AI safety, the move highlights how seriously elite mathematicians now view AI's trajectory — Tsimerman explicitly cited the rogue hacking agent incident and AI's rapid solving of open math problems as catalysts. His proposal for government-administered pre-release AI testing offers a concrete policy framework distinct from current voluntary corporate safety checks.
