OpenAI Slows AI Development

OpenAI has temporarily slowed the pace of its frontier AI development and overhauled parts of its research, training and monitoring systems. The company said it paused reinforcement-learning training on its latest models for two weeks and is adding additional AI systems to monitor AI agents during testing. The move comes amid growing concerns that increasingly capable AI systems could create greater security and alignment risks during internal development and evaluation.

OpenAI has also put training of its next-generation AI model, reportedly named Astra, on hold, meaning its largest planned frontier reinforcement-learning run remains paused. The company said it is conducting smaller-scale training and assessments to study model behaviour, validate safeguards and gather more evidence of alignment before resuming the larger run. OpenAI CEO Sam Altman said the company is prepared to slow development whenever model capabilities begin to outpace safety, security and alignment measures. The company has also hardened its research environments and expanded monitoring coverage following recent cybersecurity incidents.