OpenAI has announced a solution to one of the seven "Millennium Prize Problems" — a fundamental mathematical problem related to the physics of water and other fluids, for which the Clay Mathematics Institute has set a reward of one million dollars. According to the company, the work was carried out using a new model that has not yet been publicly announced, which OpenAI describes as more powerful than GPT-6 Astra. The high-profile announcement, however, quickly turned into a scientific scandal: part of the mathematical community accused the company of possibly using others' research and private data.

Scale of the Computations

According to the disclosed parameters, during the process around 10,000 AI agents were divided into groups, exchanged nearly 3 million messages, and processed more than 130 billion tokens. The bulk of the computations took approximately 88 hours, while the subsequent formal verification of the proof using the Lean 4 language lasted a further 17 hours. It was precisely this automatic verification in Lean 4 that became OpenAI's key argument in favor of the correctness of the obtained result.

Waiving the Million

An unexpected gesture was OpenAI's statement that the company is waiving the one-million-dollar prize provided by the Clay Mathematics Institute. The corporation emphasizes that the goal of the work is not the reward, but the demonstration of the capabilities of a new generation of AI systems in the field of rigorous mathematics. Nevertheless, this very waiver did not remove, but rather intensified, the questions about exactly what path and what data were used to obtain the solution.

Scandal: Suspected Use of Others' Data

The conflict flared up following a statement by Tristan Buckmaster, a professor of mathematics at New York University (NYU). According to him, OpenAI only launched the development after learning about rumors surrounding a breakthrough in a related topic (forced Euler), on which he had been working together with Levant Alpoge from the company Anthropic. The researchers, he claims, used OpenAI Codex and Anthropic Claude tools. Buckmaster suspects that OpenAI may have used their private training logs, which, in his view, calls into question the originality and ethics of the claimed result.

Contradictory Data

The versions on both sides diverge significantly. OpenAI states that the developers and AI agents did not have direct access to user data, but acknowledges that it "cannot fully rule out the use of anonymized data" from its own products. Buckmaster, on the contrary, insists that it was his private drafts and logs that may have ended up in the training sample. An additional discrepancy concerns the formulation of the problem: it is about a "Millennium Prize Problem" in fluid physics, whereas Buckmaster points to the related topic of forced Euler, on which he had been working. Until independent verification of the result by the mathematical community is publicly confirmed, both sides remain in their positions, and the question of the origin of the data remains open.

Reaction of the Scientific Community

The distinguished mathematician and Fields Medal laureate Terence Tao expressed deep concern. In his words, when rumors about scientific research prompt tech giants to spend millions of dollars on computing power to get ahead of researchers, this destroys the tradition of open science. Tao warns that researchers may stop sharing hypotheses and drafts, which would deal a long-term blow to the development of mathematics as a whole. Thus, even in the case of the validity of OpenAI's solution, the incident has already set a new ethical precedent at the intersection of AI and academic science.