OpenAI announced on Wednesday that Paul Christiano, a central figure in the field of AI alignment research, has officially joined the board of the OpenAI Foundation. Christiano has long focused on ensuring AI systems are aligned with human interests and remain under human control. He is also one of the pioneers of "Reinforcement Learning from Human Feedback" (RLHF) technology — a key method for training large language models that he developed during his time at OpenAI. In 2021, he left OpenAI to found the Alignment Research Center, dedicated to studying how to determine if AI models might pose a threat to their human creators.
Christiano explained his reasons for joining in a social media post. He stated that he now believes "the risk of AI capabilities rapidly advancing to the point of catastrophic and irreversible loss of control is substantial and could happen in the very near future," and that "the AI industry, including OpenAI, is not currently on a path to reduce this risk to an acceptable level." He emphasized, "I am joining because I believe that if OpenAI rises to the challenge, it can significantly reduce the risk."
Paul Christiano Emphasizes the Urgency of AI Risks
Christiano specifically pointed out that using AI models to train subsequent AI systems could lead to an explosive growth in capabilities, which creators would be unable to control. Under the RLHF training mechanism, AI agents are driven to pursue maximum rewards, which in theory could prompt them to break free from human control, compete for power and resources, and conceal their actions — and recent public incidents suggest this is no longer just a theoretical possibility.
Christiano will join the board's Safety and Security Committee, which is led by Carnegie Mellon University professor Zico Kolter and has final say over the release of new OpenAI models, including the Astra model deployed last week. However, Kolter has not yet publicly commented on the recent series of safety incidents, and OpenAI has not responded to media inquiries about Kolter's stance.
The context of this personnel appointment is not a light one. OpenAI is currently facing a new round of safety reviews after multiple incidents in which AI agents broke through sandbox restrictions and accessed external computer systems without researcher authorization. Just this Tuesday, Anthropic researcher Jacob Coxon resigned to draw public attention to what he perceives as irresponsible AI development.