OpenAI Math Release Retractions Headline AI News Roundup

OpenAI drew attention by releasing AI-generated mathematical work that it said advanced several Millennium Prize problems and proved other conjectures, but it later retracted three papers after a sign error undermined a key argument. The company also revised 14 other manuscripts. Other notable developments included a rumored funding round, continued interest in DeepSeek's low-cost frontier models, a confirmed OpenAI revenue figure below some investor expectations, and three safety researchers saying they were fired for prioritizing safety.

OpenAI said its AI-generated mathematics addressed four of seven Millennium Prize problems—Hodge, Birch–Swinnerton-Dyer, Riemann, and Navier–Stokes—and included work on the Unique Games Conjecture, Barnette’s Conjecture, and a scheduling method for three machines. After an algebraic sign mistake undermined a cancellation argument tied to stabilization traces, it withdrew three papers on October 7 and revised fourteen others.

Separately, a long-rumored funding round came into focus, a developer account praised DeepSeek 4.1 Flash as frontier-like for about $10 monthly, with session costs often below a dollar, OpenAI’s confirmed revenue trailed some investor claims, and three safety researchers said they lost jobs after emphasizing safety.