OpenAI posted that it temporarily paused reinforcement learning training on its latest models intended for deployment. The company said more capable models bring greater risks during internal development and testing, so it hardened and red-teamed research environments while expanding monitoring coverage. Sam Altman stated the pause covers some frontier RL runs to meet alignment, security, and monitoring standards and that it affects further-out releases. He added the company still expects to ship new models soon. Greg Brockman noted the slowdown includes the largest planned frontier RL effort.