OpenAI has claimed a major mathematical breakthrough, but competing claims now cloud it. The company says its AI solved the Navier-Stokes equations problem. This is one of seven famous Millennium Prize Problems in mathematics. Each carries a $1 million award from the Clay Mathematics Institute.
The problem has remained unresolved for roughly 90 years. It asks whether smooth three-dimensional fluid motion can break down mathematically. OpenAI announced on September 8 that its internal model proved this. Specifically, the equations can “blow up,” reaching impossible infinite speeds.
The scale of the effort was remarkable by any measure. OpenAI used around 10,000 concurrent AI agents on the problem. These agents reportedly reached a resolution within just 88 hours. The GPT-6 Astra model then verified it using the Lean language.
However, the announcement quickly became mired in serious controversy. NYU professor Tristan Buckmaster raised pointed questions about the work. He collaborated with Anthropic researcher Levent Alpöge on related problems. The pair had been working on the connected Euler equations.
Buckmaster alleged OpenAI learned of their unpublished progress last week. He claimed the pair shared drafts inside OpenAI’s Codex tool. He also said they discussed their approach with an OpenAI researcher. Consequently, he suggested OpenAI adopted their method for the full proof.
The allegations grew considerably more serious from there. Buckmaster claimed OpenAI proposed a joint release excluding Alpöge. Most strikingly, he alleged OpenAI scientists threatened his career. He said he asked about training data but received no answer.
OpenAI has firmly denied all these allegations as false. Researcher Sébastien Bubeck led the press briefing defending the work. He insisted the internal model solved the Euler problem independently. Furthermore, he said no user prompts or proofs were accessed.
However, OpenAI acknowledged one important nuance in its statement. Its full Navier-Stokes solution followed a similar method to the pair’s. Additionally, it admitted de-identified usage data might have helped models. The company said it recognizes the priority of their Euler work.
