OpenAI told the New York Times it is working through how to share another set of mathematics results. The company said the work reflects substantial progress on another Millennium Prize problem, following its disclosure that an internal model had produced what OpenAI described as a Navier-Stokes proof.

Earlier reports named the Hodge Conjecture as close to verification by OpenAI and said either OpenAI or Anthropic may also be nearing Birch-Swinnerton-Dyer, but those claims were presented as rumors. If verified, that would amount to 3 Millennium problems in a month. OpenAI had previously said it assigned separate agent groups to all open Millennium Prize problems, with the Navier-Stokes effort alone using 10,000 agents.

Separately, Andreas Thom, a group theorist at TU Dresden, questioned whether unpublished work he discussed with ChatGPT may have helped train OpenAI's Astra model. Thom said researchers told him his conversations had not entered training data, but he pointed to OpenAI's later statement in a separate Navier-Stokes dispute that it could not rule out de-identified user data improving its models. His comments are the second public accusation by a mathematician in a month.