Terence Tao argues AI racing to solve open problems threatens open science traditions
OpenAI agents claim Navier-Stokes Millennium Prize proof amid dispute
Formal AI resolution of a Millennium Prize Problem, if the proof stands, would be a significant capability advance. The dispute over priority and possible access to unpublished research raises unresolved questions about how AI development interacts with academic research norms.
The full picture
An OpenAI multi-agent system, running roughly 10,000 concurrent agents on an internal model described as significantly more capable than GPT-6 Astra, produced and Lean-verified a proof that three-dimensional Navier-Stokes equations develop a finite-time singularity. Agents were launched September 1, 2026; the proof was completed approximately 88 hours later on September 5, with Lean formalization taking an additional 17 hours. The result is disputed by mathematicians Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic), who had independently achieved a Lean-verified proof for the related 3D Euler equations by August 22 and had been working on the broader problem for nearly a year before OpenAI launched its effort. They raise the possibility that OpenAI's effort benefited from knowledge of their unpublished work. Mathematician Terence Tao has separately argued that AI systems racing to solve open problems upon hearing rumors of ongoing research may push mathematicians toward secrecy, threatening open science traditions.
How it developed
OpenAI publishes proof, manuscripts, and Lean certificates alongside public announcement
Sources
1 more source
Related
- Grew out ofOpenAI's automated research intern milestone
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free