NYU mathematics professor Tristan Buckmaster announced three proofs this week, contributing to a major unsolved problem in theoretical mathematics. Buckmaster collaborated with Anthropic mathematician Levent Alpöge on the work, using both Codex and Claude AI models throughout.
There is another part of this story,
Buckmaster wrote in his public statement, expressing regret at needing to address the issue at all. His statement detailed a parallel effort by OpenAI that allegedly built on his team’s work before it was ever released publicly.
A Race to a Million Dollar Proof
The Navier Stokes existence and smoothness problem is one of seven Millennium Prize problems, with the Clay Mathematics Institute offering $1 million for its solution. The equations are fundamental to fluid mechanics, and a proof would mark a significant advance in mathematical physics.
Shortly after Buckmaster’s statement went public, OpenAI published a full proof of its own. The company claims an unreleased next generation model discovered the proof, reportedly tackling several unsolved problems over the past week. The effort consumed 300 billion output tokens, valued at $22.5 million at current Astra rates.
Questions OpenAI Wouldn’t Answer Directly
Buckmaster and Alpöge were finalizing their results when they learned their progress had reached OpenAI. When contacted, OpenAI stated they already had a full proof, but questions about their research start date and how much human input was involved received evasive answers.
It emerged that an entire team had been working on the problem,
Buckmaster said, noting an insane amount of compute
had been thrown at it. OpenAI eventually admitted their initial prompt was sent in the past few days, and only after information about Buckmaster’s work had already surfaced.
A Suspicious Coincidence in Approach
Buckmaster found it telling that OpenAI adopted the same specific, uncommon tactic his team had used. Almost nobody else I know of was working on it,
he wrote. OpenAI’s own post confirms their latest effort began September 1st, inspired by rumors that Millennium Prize problems were being solved.
Buckmaster also alleges that an OpenAI researcher asked him to remove Alpöge’s credit from the work, and claims the same researcher then threatened his career when he refused.
The Codex Question Nobody Can Fully Answer
Buckmaster used OpenAI’s Codex model extensively throughout his project. OpenAI reserves the right to train its models on Codex interactions, raising the possibility that his own work could have indirectly informed OpenAI’s parallel effort.
- OpenAI denies accessing Buckmaster’s specific user data
- The company cannot rule out de-identified data contributing to model improvements
- Buckmaster believes public transparency is the only path forward from here
The incident is likely to intensify debates over AI’s growing role in mathematical research, and whether companies building the tools researchers rely on face a structural conflict of interest when those same tools feed data back into their own competing efforts.
Hashlytics Take
The actual scandal here isn’t that OpenAI solved a hard problem fast. It’s that the fastest path to solving it may have run directly through the tool the original researcher was using to do his own work. If a company’s coding assistant can plausibly absorb a researcher’s unpublished progress and that same company then publishes a competing result days later, the incentive structure is broken regardless of what actually happened internally. Denying specific data access while admitting de-identified data “cannot be ruled out” is not the reassurance OpenAI seems to think it is.
Follow Hashlytics on Bluesky, Facebook, LinkedIn , Telegram and X to Get Instant Updates
