Math world erupts over claims OpenAI used private Codex sessions for Navier-Stokes

A Twitter thread from user Teortaxes has lit up the math community with mutual accusations between leading researchers and OpenAI over a possible Navier–Stokes solution. At the center: Tristan Buckmaster and Levent Alpöge, who allegedly claim OpenAI tapped their private Codex sessions to reach a result using a program by Córdoba and Martínez-Zoroa.

There's history, though. Teortaxes says Alpöge spent about two months clashing with mathematicians over what's called "unorthodox communication." That background initially tilts sympathy toward OpenAI — at least for the outside observer. The thread itself is a thicket of hedges: "if I'm reading this right," "maybe not," "maybe the opposite." Good luck pinning down a solid fact.

The core allegation: OpenAI cracked Navier–Stokes by training on Buckmaster and Alpöge's private Codex sessions. Teortaxes doubts OpenAI trains on Codex data "that fast" and offers an alternative read: OpenAI offered coauthorship in exchange for picking Codex, otherwise "go to hell." Meanwhile, Anthropic — "Ant" in the thread — is allegedly trying to hijack the greatest mathematical achievement of the century.

What's actually clear? Nothing. No official statement from OpenAI, Buckmaster, Alpöge, or Anthropic. No technical details on the Córdoba and Martínez-Zoroa program. No dates. No evidence private sessions were used. No confirmation Navier–Stokes was actually solved. Teortaxes sums it up: either way, at least one trillion-dollar AGI company takes another reputation hit. Until someone goes on the record, this is social-media speculation, not a verified story.