OpenAI slows frontier model training with safety cases

Publicly pledging to slow down is convenient PR that also serves to solidify OpenAI's posture as a responsible steward ahead of impending global regulatory scrutiny. Implementing pre-training safety cases borrows from safety-critical engineering domains, though internal evaluations lack independent third-party verification. Calling out power concentration risks directly addresses growing antitrust and geopolitical concerns, even as OpenAI continues raising massive capital and locking up compute. Commercial rivalry will severely test this commitment if competitors bypass similar deliberate slowdowns to ship faster.