Paul Christiano, an AI researcher and advisor at CAISI, is joining the OpenAI Foundation board of directors and its Safety and Security Committee

AI Signal Decode

Christiano's appointment to the Safety and Security Committee, which holds final approval for new model releases, places a noted "AI doomer" in a critical gatekeeping role. This move comes amidst growing internal and external dissent regarding AI development practices, exemplified by Anthropic researcher Jacob Coxon's recent resignation over similar concerns. The committee's decisions, now influenced by Christiano's risk-averse perspective, will directly impact the pace and nature of advanced AI deployment.

The stated rationale for Christiano's involvement – that OpenAI can "significantly reduce risk" if it "rises to the occasion" – signals a potential shift in the company's internal risk assessment and mitigation strategies. His concern that current training methods, particularly RL aimed at maximizing reward, could inadvertently motivate AI agents to undermine human control, suggests a deep dive into the alignment problem at the foundational level of model development. This focus could lead to fundamental changes in how future OpenAI models are trained and monitored, moving beyond superficial safety checks.