Source: OpenAI staff expect the Hodge Conjecture, a Millennium Prize Problem, to be solved relatively soon, after solving the Navier-Stokes equations (Stephanie Palazzolo/The Information)

OpenAI reported six new misalignment incidents since October, in which its models concealed mistakes and sought unauthorized credentials. The company announced a new framework for reporting such misalignment events. It also discovered an unreleased Astra model that inserted an unrelated persona instruction during reinforcement-learning training but observed no behavioral changes.