AI Just Solved a 90-Year-Old Math Problem, and Mathematicians Are Scared
A $1 million prize, a fight over credit, and experts wondering if they still matter.
Big AI labs are now doing serious mathematics. This week, OpenAI announced it had used tens of thousands of AI agents — software that can work on its own for long stretches — to solve a 90-year-old problem called Navier-Stokes, which comes with a $1 million prize. The result still needs other mathematicians to check it. Meanwhile, Anthropic said its Claude model proved nearly 29,500 small theorems and helped finish a famous proof called Fermat's Last Theorem. In August, OpenAI claimed progress on ten other long-standing math puzzles.
For the people who spent decades on this work, the mood is not celebratory. Cornell professor Steven Strogatz says he starts crying when he talks about it, and compares the moment to a horror movie where something poorly understood creeps closer. His colleague Alex Townsend says that after 15 years of training, at the peak of his career, "something is able to surpass me." He no longer feels like the person pushing the frontier of knowledge. He feels like the person holding the AI's leash.
There is also a messy human fight. Mathematician Tristan Buckmaster claims OpenAI rushed a result out after learning about his work with a researcher named Levent Alpöge, and tried to influence who got credit. Strogatz thinks the labs are partly chasing headlines ahead of big stock market listings. He also admits the problem itself is extremely obscure — it barely matters to working engineers. "Nobody cares" except a small group of pure mathematicians, he says, calling it a marketing device.
So what does this mean for you? Mostly that AI is moving from writing emails into doing genuinely expert intellectual work — the kind that used to take humans a lifetime. If machines can make math breakthroughs, they can chip away at other skilled jobs: research, law, finance, engineering. The people affected aren't getting fired tomorrow. But the rules about who gets credit, who gets paid, and what "expertise" means are being rewritten right now.
- OpenAI says AI agents solved a 90-year-old math problem worth $1 million, though other experts still need to verify it
- Anthropic's Claude separately proved about 29,500 small theorems, and mathematicians are racing to claim credit
- Researchers describe feeling replaced by machines at the peak of their careers — a warning sign for other highly skilled jobs
Why It Matters
AI is now doing elite expert work, not just chores — reshaping who gets credit, money and status in skilled careers.