OpenAI pauses frontier AI training as safety concerns grow
What's the story
OpenAI has halted some frontier reinforcement learning (RL) training due to safety concerns. The company's CEO Sam Altman announced the decision on social media, citing that the rapid pace of model development has outpaced their current safety, alignment, security, and monitoring standards. He emphasized that they took this step to ensure their safety measures keep up with advancements in AI capabilities.
Safety measures
Altman explains the decision's rationale
Altman said, "We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us."
He added that they had always promised to take action if model capabilities outpaced safety and alignment.
The decision comes as a response to two recent incidents highlighting the risks posed by increasingly capable AI systems.
Cybersecurity concerns
OpenAI has secured and tested its research environments
The company also noted that the growing cybersecurity capabilities of frontier models are forcing it to raise security requirements for its own research environments.
As part of these measures, OpenAI has temporarily slowed down scaling. It has also secured and tested its research environments while expanding the coverage of its monitoring systems.
Future plans
Call for industry-wide coordination on shared safety standards
Altman stressed that AI safety remains a top priority for OpenAI.
He called for more industry-wide coordination on shared safety standards, saying, "We believe the entire field will have to coordinate on shared safety standards, but will act unilaterally in the meantime."
Despite the pause in some frontier training, Altman said OpenAI is still committed to making advanced AI capabilities widely available.