OpenAI accused of dirty tactics on $1 million math problem

The controversy centers on a clash between traditional academic norms and the immense compute power available to frontier AI labs. Buckmaster and Alpöge chose an uncommon, specific route to attack the Navier-Stokes problem, one that he says few others were pursuing. When OpenAI announced a full proof shortly after his preliminary findings, he found it suspicious that they had arrived at the same niche approach.

Buckmaster's concern is amplified by the fact that he used OpenAI's Codex tool extensively. Since OpenAI reserves the right to train models on Codex interactions, he worries his private work could have been regurgitated. OpenAI's response downplays this, asserting researchers and agents saw no specific user data, but concedes it cannot rule out that de-identified data helped improve their models.

The outcome may hinge on the credibility of OpenAI's claim that its effort began on September 1, inspired by rumors of solved problems, rather than by Buckmaster's work. For the mathematical community, the episode could define how research credit is assigned when AI systems are involved and whether using a lab's tools creates an implicit leak of one's own ideas.