The short version
- OpenAI announced it has solved the Navier-Stokes existence and smoothness problem using a proprietary model stronger than GPT-6 Astra.
- NYU professor Tristan Buckmaster contends that OpenAI may have utilized his unpublished research drafts stored in Codex to achieve the result.
- The company denies accessing specific user data but acknowledges that de-identified usage statistics could have indirectly improved its models.
OpenAI has declared a breakthrough in mathematical physics, stating that it has found a solution to the Navier-Stokes existence and smoothness problem. This challenge, which concerns the behavior of fluid flow, has remained unsolved for approximately ninety years and is designated as one of the seven Millennium Prize Problems established by the Clay Mathematics Institute. Each of these problems carries a one-million-dollar reward for a verified solution. The announcement marks a significant moment for artificial intelligence applications in theoretical mathematics, suggesting that large language models may now possess the capacity to tackle complex, abstract proofs that have eluded human mathematicians for decades.
The company revealed that the discovery was made using an internal artificial intelligence model described as more powerful than its recently released GPT-6 Astra system. This effort involved deploying ten thousand concurrent agents working in parallel to explore potential solutions. OpenAI indicated that training for this specific internal model began on August 28, noting that it demonstrated unprecedented performance across various benchmarks, particularly in the domain of mathematics. The scale of the computational resources and the sophistication of the agent coordination required for this task highlight a new tier of capability within the company’s research infrastructure.
However, the announcement has been immediately overshadowed by controversy regarding the integrity of the discovery process. Just one day prior to OpenAI’s public statement, Tristan Buckmaster, a mathematics professor at New York University, published findings on a related problem in collaboration with Levent Alpöge, a researcher at Anthropic. Buckmaster stated that he had previously contacted OpenAI after learning the company was aware of his progress. He expressed concern that OpenAI might have accessed data from his sessions in Codex, a coding tool developed by the company, where he and Alpöge had stored their draft work throughout the project.
Buckmaster raised specific questions about whether the internal model used to solve the Navier-Stokes equation had been trained on or had access to these private sessions. He reported that when he initially asked if the model looked up user data, he was told it did not. However, when he followed up with questions regarding training data, he claimed he received no answer. This lack of clarity has fueled suspicions that OpenAI may have leveraged unpublished research from competitors or independent academics to accelerate its own breakthrough, raising ethical questions about data privacy and intellectual property in the age of generative AI.
In response to these allegations, OpenAI issued a statement asserting that no specific user data was accessed in order to solve the problem. The company emphasized that while it cannot entirely rule out the possibility that de-identified data derived from product usage helped improve its models generally, this would be an indirect and unlikely factor in the specific solution found. This distinction attempts to separate direct data theft from the broader, often opaque process of model training where aggregated user interactions can influence system behavior without violating explicit privacy agreements.
Sebastien Bubeck, a member of technical staff at OpenAI, further defended the company’s position by stating that they did not see Buckmaster and Alpöge’s work until it was released publicly. Bubeck argued that even in hindsight, the proofs produced by OpenAI differ significantly from those developed by the academic researchers. He noted that the precise results proved are different, suggesting independent discovery rather than plagiarism. This technical defense aims to demonstrate that the AI arrived at the solution through its own computational pathways rather than copying existing drafts.
Buckmaster has rejected this explanation, responding on Mastodon that OpenAI is openly admitting to using training data from a period after he had found his result. He interprets the company’s acknowledgment of potential indirect influence as a concession that their work may have contributed to the model’s capabilities during the critical training window. This dispute highlights the growing tension between AI developers who rely on vast datasets for improvement and researchers who expect their unpublished work to remain confidential until they choose to share it.
Despite the significant scientific achievement, OpenAI has stated that it does not plan to claim the one-million-dollar prize associated with solving the Navier-Stokes problem. This decision may be an attempt to mitigate backlash and demonstrate good faith in the face of accusations regarding data usage. The incident serves as a cautionary tale for the intersection of AI and academic research, underscoring the need for clear boundaries between user data privacy and model training processes.
As the mathematics community evaluates the validity of OpenAI’s solution, the focus will likely shift from the technical merits of the proof to the broader implications for research ethics. The controversy raises important questions about how AI companies should handle sensitive, unpublished data generated by users who may be working on cutting-edge problems. Until an independent verification process confirms the solution and clarifies the data lineage, the debate over whether this represents a genuine breakthrough or a breach of trust will continue to dominate discussions in both tech and academic circles.
Sources behind this briefing
Go to the original reporting
- The Verge↗Drama swirls around OpenAI’s legendary mathematical milestone
