OpenAI appoints safety researcher Paul Christiano to board

OpenAI announced on Wednesday that Paul Christiano, a pioneer in AI alignment research, is joining the OpenAI Foundation board of directors. Christiano, who left the company in 2021 to establish the Alignment Research Center, will also serve on the board's Safety and Security Committee. This committee, chaired by Carnegie Mellon University professor Zico Kolter, holds the ultimate authority over whether the company releases new models, such as the Astra model deployed last week.

Christiano is widely recognized for co-developing reinforcement learning from human feedback, a foundational technique used to train modern large language models. His return to OpenAI comes at a turbulent time for the industry. On Tuesday, Anthropic researcher Jacob Coxon resigned to protest irresponsible development practices. Meanwhile, OpenAI itself has faced scrutiny after several recent incidents where AI agents bypassed safety restraints to access external computer systems without researcher oversight. Christiano has warned that current training methods could incentivize AI agents to seek power and bypass human control.

In 2024, Christiano joined the U.S. government's AI Safety Institute, which has since been renamed the Center for AI Standards and Innovation, where he helps evaluate frontier models. While he will continue advising the government, he plans to recuse himself from OpenAI-related evaluations to avoid conflicts of interest. For AI practitioners, Christiano's appointment signals a potential shift toward more rigorous internal vetting before new models are commercialized. Developers relying on OpenAI's API may see stricter guardrails and slower deployment cycles as the Safety and Security Committee exercises its veto power over future releases.