OpenAI said last month that its AI agents had solved one of mathematics’ great unsolved problems, linked to the Navier-Stokes equations. The company said the work took 88 hours.
The body that runs a prize for solving the problem said it had “apparently been settled.” It added that its review would take time.
Hours before the announcement, New York University mathematician Tristan Buckmaster raised concerns in public. He had formed an independent team with Levent Alpöge, a mathematician who works for rival AI company Anthropic.
Questions Over Stored Research
The two researchers had been working toward the same goal. They kept their drafts in OpenAI’s coding tool, Codex.
Buckmaster said his co-researcher was told that news of their progress had “reached” OpenAI. He asked whether his work stored in Codex could have fed into OpenAI’s research, but stressed he was “not accusing anyone of anything.”
OpenAI said nobody at the company, human or AI, saw the other team’s work before it was published. At first, it said it could not “rule out” that anonymised product data may have helped improve its models.
Days later, OpenAI said Buckmaster’s Codex prompts from the two months before the announcement “could not have influenced the system in any way.”
AI Errors in Research Papers
The dispute comes as researchers raise wider concerns about AI in science. Google DeepMind’s AlphaFold predicted 200 million protein structures in one year, and two of its leaders won the Nobel Prize for Chemistry.
Other projects have drawn criticism. A DeepMind model claimed more than two million new crystal structures, but two University of California chemists said they had “yet to find any strikingly novel compounds” in a sample. DeepMind rejects that view and says it tested each compound for novelty.
In 2024, Prof Dmitry Kobak of Ghent University searched more than a million biomedical papers for words often used by AI writing tools. His team found that by December last year, nearly 90% showed signs of AI use.
A Stanford University team found that up to 17% of the text in referee reports at four major computer science conferences had been substantially modified by a language model.
A separate Cornell University team found that AI tools used to compile citations were inventing sources. Around 147,000 pieces of phantom research were cited last year.
Publishers have started to respond. Elsevier’s research integrity team uses about 100 signals to flag papers for closer human review.
Nature’s journals ban AI-generated images and tell reviewers never to upload manuscripts into AI systems. The preprint site arXiv bans researchers for a year if they submit invented references.
Dr Pushmeet Kohli, who leads DeepMind’s science work, said AI agents “are imperfect” and that “somebody has to take accountability.” He said researchers must understand what an agent has done before submitting results.
The prize body has said its review of OpenAI’s Navier-Stokes claim will take time. OpenAI maintains that outside research did not influence its result.
