OpenAI's math claim ignites dispute over AI data privacy

OpenAI announced last month that its AI agents solved a major unsolved problem related to the Navier-Stokes equations in 88 hours. The International Mathematical Union, which administers the prize for this specific problem, stated that the issue has "apparently been settled" but noted that a full review will take time. The company’s claim of a rapid breakthrough has drawn immediate scrutiny from the academic community regarding the data sources behind the solution.

Tristan Buckmaster, a mathematician at New York University, raised concerns hours before OpenAI’s announcement. He formed an independent team with Levent Alpöge, a mathematician employed by Anthropic, to work on the same problem. Buckmaster said his co-researcher was tipped off that news of their progress had reached OpenAI. He asked whether his drafts, which were stored in OpenAI's coding tool Codex, could have fed into the company's models, though he stressed he was "not accusing anyone of anything".

OpenAI responded that no human or AI at the company saw the other teams' work before it was published. Initially, the firm said it could not "rule out" that anonymised data from product use may have improved its models. Days later, OpenAI clarified that Buckmaster's Codex prompts from the two months prior to the announcement "could not have influenced the system in any way".

The dispute is part of a wider concern about AI undermining scientific trust. Researchers worry that AI chatbots are increasingly hallucinating facts in scientific contexts. If these errors become common, it may become difficult to distinguish true discovery from AI-generated fiction. This threatens the core of scientific research: the reliability of published findings.

Dalmeet Singh Chawla, a science journalist, stated that generative AI has disrupted the scientific process. He noted that publishers' integrity staff, who screen submissions for accuracy, will need to drastically increase in number and effort to keep up. He observed that currently, those staff members seem unprepared for the volume of AI-generated distortions arriving at journals.

Evidence of this strain is visible in recent studies. Professor Dmitry Kobak of Ghent University found that nearly 90% of a sample of biomedical research papers had AI fingerprints by December 2025. Separately, researchers at Cornell University found that AI tools used to compile citations were inventing bogus research, with around 147,000 pieces of phantom research cited last year. These phantom citations can misdirect funding and slow valid scientific progress.

Journals are implementing new defenses to maintain integrity. Elsevier has developed approximately 100 signals to flag papers for closer human scrutiny. Nature’s journals forbid AI-generated images and instruct referees not to upload manuscripts into AI systems. The preprint server arXiv now bans researchers for one year if they submit invented references. These measures aim to ensure that human scientists remain in control of the research process.

Pushmeet Kohli, who leads science work at Google DeepMind, compared abandoning AI to computing without calculators. He argued that AI agents are currently imperfect and that humans must take accountability. He stated that researchers must understand what an agent has done and cannot simply let a machine write and submit a paper as true without human verification.