OpenAI and mathematician dispute raises questions about AI research norms

In short: A dispute over who solved a famous maths problem first is fueling a wider debate about how AI should be used in research.

What's going on

The Navier-Stokes problem is a long-running maths challenge about how fluids move, like air around a plane wing or water in a pipe. It has been unsolved since 1934 and is part of the Millennium Prize Problems, a set of seven questions with a $1 million prize each.

The Financial Times reports that two mathematicians, Tristan Buckmaster (New York University) and Levent Alpöge (Anthropic), believed they had solved Navier-Stokes in August after about a year of work. They used large language models, including Anthropic’s Claude and OpenAI’s Codex, along the way. Soon after, OpenAI said it built a new internal maths-focused AI model, then ran an intensive push and announced its own solution in early September.

Buckmaster said OpenAI appeared to learn about his team’s progress and then pursued the same approach, which he described as unusual. He raised a concern about whether his use of Codex could have exposed information that helped OpenAI, though he did not claim proof. OpenAI said its researchers and systems did not access his Codex work or usage data, and that his conversations could not have been used to train the model.

What to watch

Even if no rules were broken, the episode is prompting hard questions. Researchers worry that sharing early ideas could become risky if a well-funded lab can sprint to the finish and announce first. Others argue that fast solutions still matter, especially in areas tied to medicine or public safety. The next signs to watch are whether academics share less, whether companies keep AI breakthroughs quiet, and whether more groups move to locally run models (like keeping your notes in a locked desk instead of a shared office system).

Source: Financial Times