OpenAI claims Navier-Stokes solution in 88 hours with 10,000 AI agents

OpenAI says it solved the 90-year-old Navier-Stokes Millennium Prize Problem in 88 hours using a system of roughly 10,000 concurrent AI agents. The claim is staggering, but independent verification is pending, and a dispute over data provenance has already erupted. For AI builders, the episode reveals both the potential and the unresolved trust issues in AI-assisted research workflows.

How OpenAI describes the agent architecture

OpenAI said it deployed a system of "coordinating agents" powered by an internal model that is not released to the public source: CNBC. The agents had tools including the ability to read from a cached version of the internet and run code. They were subdivided into groups with intra-group communication, and the group that produced the Navier-Stokes resolution involved on the order of 10,000 concurrent agents. The agents reportedly arrived at their resolution on September 5, about 88 hours after the first agents were launched. OpenAI also published Lean proofs alongside the release, though the formal proof is not public source: CNN.

The data provenance dispute

The night before OpenAI's announcement, NYU mathematician Tristan Buckmaster released a statement saying he and Anthropic researcher Levent Alpöge had been working on the same problem using large language models, including OpenAI's Codex source: CNN. Buckmaster said Alpöge received "tips" that information about their progress had been passed to OpenAI, and he noted that OpenAI's route was similar to their own work, "not the direction one arrives at in a few days by giving a model the problem statement" source: CNBC.

OpenAI said its effort began on September 1 after hearing a rumor about progress on the puzzle, and that its agents did not see any of Buckmaster and Alpöge's work through any means until they released it publicly. The company stated that "no specific user data was accessed in order to solve this problem," but added that "while unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models" source: CNBC.

Why builders should watch the verification pipeline

For teams building AI agents that produce research outputs, this case highlights three open problems. First, the output of an agent swarm is not self-validating. Even with a Lean proof, the mathematical community will demand independent review, and no one has verified the solution yet. Second, data provenance matters when AI systems train on user activity. If your agents interact with proprietary data, you need clear auditing. Third, OpenAI stated it does not intend to claim the Millennium Prize source: CNN, which removes some pressure but also means the claim remains an unreviewed assertion.

Caveats and uncertainties

The Clay Mathematics Institute has not commented on OpenAI's proposed solution. Mathematician Terence Tao warned that using AI to jump directly to a solution risks flattening the field: "it is a little like going to watch a movie and jumping straight from the first ten minutes to the last ten minutes; technically, all the plot lines are resolved, but most of the value of the experience was lost" source: CNN.

Until the proof is publicly released and independently checked, this remains a vendor claim, not an established fact. The dispute over data influence may take months to resolve. For now, the most practical takeaway is that multi-agent systems can explore difficult mathematical terrain quickly, but trust and verification infrastructure still lag behind the raw capability.