OpenAI Accused of Using AI Agents to Produce Math Proof in 88 Hours

A New York University mathematician and an Anthropic researcher, working in their personal capacities, spent nearly a year studying problems related to the Euler and Navier-Stokes equations, using AI tools for computations and formal verification. Tristan Buckmaster and Levent Alperg had planned to keep polishing their paper before its progress circulated on social media. OpenAI then reportedly mobilized a large number of AI agents to focus on the same direction, and its model produced a proof within 88 hours. Both sides subsequently released their results, sparking debate.

Buckmaster publicly questioned the way OpenAI intervened, saying his core concern was not that OpenAI stole data, but that a commercial company, after learning of the research buzz, used massive computing power to rush at the same problem. OpenAI denied accessing the researchers' paper before its release and denied accessing their private data.

The result concerns the Navier-Stokes equations under specific conditions and does not amount to a full solution to the Millennium Prize Problem. Formal verification also cannot replace peer review. The dispute went beyond authorship. Several mathematicians, including Terence Tao, issued public statements saying they do not oppose using AI in mathematics, but oppose treating hard problems as model benchmarks, rushing to publish answers while neglecting paper writing, peer review, citation and intellectual heritage.