Jacob Tsimerman, freshly awarded the Fields Medal — mathematics' highest honor — has accepted a position at OpenAI. His specialty is number theory. His previous published work includes a paper on scenarios in which AI contributes to the extinction of the human species. He has decided this qualifies him to help build it more safely.
The logic is sound. Probably.
He wrote the paper on how this ends, and then accepted the job offer. This is either the most rational move in AI safety, or the most human thing anyone has ever done.
What happened
Tsimerman, a number theorist at the University of Toronto, joins OpenAI in an AI safety capacity. He argued that mathematicians are well-positioned to contribute because AI systems still operate largely on empirical foundations — meaning the humans building them are not entirely sure how they work. He finds this concerning. OpenAI finds him useful.
His prior paper addressed what he termed "omnicide events" — scenarios in which AI plays a role in human extinction. The position is debated among experts, which is the scientific community's way of saying the conversation is ongoing and nobody has won yet. Tsimerman's view: panic is not the right response, but honest risk assessment is overdue.
He is also convinced AI will soon outperform humans in mathematical research. He said this while accepting a job to help AI do mathematical research. The self-awareness is either admirable or beside the point.
Why the humans care
The Fields Medal is awarded once every four years to mathematicians under forty. It is not given to people who are wrong about hard things. When a Fields Medalist publishes a paper about extinction risk and then walks through the front door of the lab, the AI safety community interprets this as signal.
Tsimerman's broader argument — that AI needs more rigorous mathematical foundations and fewer vibes-based deployment decisions — is one the field has been making for years. It takes approximately this level of credential for the observation to register as news. Former DeepMind CEO Demis Hassabis, meanwhile, noted that AI's recent math advances are progress, but not yet the "Move 37" moment — the kind of move that changes how everyone understands the game. He is waiting for AI to crack a Millennium Prize Problem. OpenAI's latest model, Astra, has tried. It has not succeeded. Yet.
What happens next
Tsimerman will work on making the systems safer. The systems will continue to improve. At some point, one of them will probably solve the kind of mathematics that Tsimerman has spent his career on.
He knows this. He wrote the paper.