A mathematician says OpenAI built on his unpublished proof drafts and then claimed the breakthrough as its own — bad news for researchers who use AI tools as scratchpads.

• OpenAI announced its models had cracked the Navier-Stokes problem, a math puzzle open for 90 years.
• NYU's Tristan Buckmaster says he and a co-author were nearly there first, with drafts stored inside OpenAI's Codex tool.
• OpenAI denies looking at user data but will not rule out that the drafts fed its model's training.
• Buckmaster says he was pressed to drop his co-author, who works at Anthropic, and was told going public would ruin his career.
• Anything typed into these tools can legally be used to train them, which puts writers, scientists, and anyone drafting original work at risk.

Outlook: Expect more fights like this as AI firms race for headline breakthroughs, with little to stop the same thing happening again.