OpenAI Pumps the Brakes on Frontier Model Training
OpenAI announced it's slowing down parts of its AI development pipeline to focus on security and safety review. The company confirmed a two-week pause in reinforcement learning training for models nearing deployment, plus an open-ended delay to what it calls its largest planned frontier RL training run.
The timing is notable: OpenAI is reportedly eyeing an IPO while facing mounting pressure from Anthropic, Chinese labs, and a growing field of open-weight models. Choosing to pause now, rather than ship faster, is being read as a real-world test of "voluntary pacing" — an idea safety researchers have floated for years but that few labs have actually acted on under competitive pressure.
OpenAI hasn't detailed exactly what safeguards prompted the delay, only that security and safety tightening are the reasons.