OpenAI claims its unreleased model solved the Navier-Stokes existence and smoothness problem, one of mathematics’ seven Millennium Prize problems, in just 88 hours. The company deployed a swarm of roughly 10,000 AI agents to tackle the fluid dynamics puzzle that has stumped human researchers for nearly 90 years. While OpenAI hails this as a major milestone demonstrating AI’s accelerating impact on pure mathematics, the announcement arrives wrapped in serious ethical controversy. Critics allege the tech giant rushed to publish only after learning competitors were close to a breakthrough, violating long-standing academic norms of collaboration and credit.
The tension escalated when NYU professor Tristan Buckmaster contacted OpenAI regarding his own recent progress on a related problem with Levent Alpöge, a researcher at rival firm Anthropic. Buckmaster reports that an OpenAI researcher responded with hostility, asking, “Why would you ruin your career?” when Buckmaster threatened to go public with his concerns. The exchange reportedly included a warning: “If you don’t want me to be nice, then I don’t have to be nice.” OpenAI allegedly urged Buckmaster to drop Alpöge as a co-author and credit OpenAI’s internal model instead. These allegations strike at the heart of academic integrity, raising fears that corporate AI labs may prioritize speed and proprietary dominance over transparent scholarly practice.
OpenAI denies accessing specific user data from Buckmaster’s Codex sessions but admits it cannot fully rule out indirect influence from de-identified usage data. The company states it began work on the problem only after hearing rumors of progress on Twitter, spending millions in a last-minute push without intending to claim the $1 million Clay Mathematics Institute bounty. This defensive posture leaves many questions unanswered. Why expend such resources if not for the prize or priority? The lack of clear prior public commitment to this specific problem makes the sudden surge appear reactive rather than planned.
This incident signals a troubling shift in how mathematical discovery might proceed. If large language models can indeed compress decades of human effort into days, the incentive structure for open research faces disruption. However, the opacity surrounding training data and the aggressive tactics alleged here suggest that trust, not just computational power, will determine whether AI becomes a partner or a predator in academia. The field now waits to see if the Clay Institute accepts the proof, but the reputational cost to OpenAI may already be incurred.
