The Credit Fight in the AI Era: Debate Sparked by OpenAI's Claim on the Navier–Stokes Problem

The report records OpenAI's claim that an AI model solved the Navier–Stokes problem and that the proof was verified with the programming language Lean, alongside the Clay Mathematics Institute's statement that it will consider validity only after peer-reviewed publication and further community vetting. It places an AI claim about a high-profile Millennium Prize problem next to the existing procedures by which mathematical results are recognized. Based on institutional statements relayed in the report; no proof details or independent verification are provided.

The report lays out concrete threads of the credit dispute: Buckmaster says he and Alpöge had been working on the problem with OpenAI and Anthropic tools and that the company moved in after learning of their work, while OpenAI says no user inputs past 3 July could have influenced the system and that it began work on 1 September. It moves the dispute from abstract principle to a checkable timeline and account settings, including Buckmaster's statement that only two of his three ChatGPT accounts had opted out of training permission. Based on public statements and social-media posts from both sides presented side by side; the report offers no independent finding.

The report records Andreas Thom's account that he spent over a year discussing group theory with ChatGPT using a particular strategy to construct a non-sofic group, and that OpenAI posted a preprint in August reporting the first example of such a group with a similar strategy; he says the OpenAI paper correctly cites earlier work by him and his collaborators but that it is unknown whether his sessions were used. It extends the credit question from cited published literature to unwritten discussion knowledge, the kind of trade craft mathematicians would normally acknowledge. Based on Thom's own assessment as relayed in the report and OpenAI's lack of a direct response on that specific question; the report states it is not established whether the model drew on his conversations.

The report cites an open letter from 25 Fields Medal winners saying AI tools muddy appropriate credit and raise 'severe attribution and plagiarism questions', and quotes intellectual-property scholar Luke McDonagh warning that researchers may not have fully grasped the consequences of uploading data and knowledge to a personal AI model account. It raises individual disputes to a collective concern about credit norms and training-data traceability, and flags personal account settings as an easily overlooked link. Based on the open letter and interviewed scholars' views; these are positions and warnings rather than quantitative evidence.

This summary is based on the single Nature news report, with no attached papers, so it is suited to understanding the event's thread, the points of dispute, and the public positions of those involved; the report does not provide material to judge the proof's technical content, the specific scope of the Lean verification, or whether the relevant conversations were actually in the training data. It speaks to mathematicians, research administrators, journals and funders, and readers concerned with the relationship between AI and scholarly norms.

The report states explicitly that independent confirmation of OpenAI's solution remains to be seen and that the Clay Mathematics Institute's recognition presupposes peer-reviewed publication; Buckmaster's and OpenAI's accounts of training-data influence conflict and the report offers no adjudication; whether the similarity Thom describes means the model used his conversations is described as not established. In addition, the report breaks off at 'Tutoring through chatbots', so the remainder is not included in the evidence bundle and cannot be incorporated here.