Paul Christiano Joins OpenAI's Board and Safety Committee

OpenAI announced on September 9, 2026, that Paul Christiano — co-inventor of RLHF and one of the industry's most vocal critics on AI existential risk — is joining the OpenAI Foundation board and its Safety and Security Committee, the body with final say on launches like GPT-6 Astra.

Who Christiano is, and why the appointment stands out

Christiano spent several years on OpenAI's alignment team, where he helped develop RLHF (reinforcement learning from human feedback), the technique that made it possible to train models like ChatGPT to follow instructions safely. He later founded the Alignment Research Center (ARC) and has become one of the public figures most closely associated with concern over catastrophic risks from advanced AI — the press routinely describes him as one of the sector's most prominent "doomers." He currently serves as a senior technical advisor at the Center for AI Standards and Innovation (CAISI), inside NIST (the National Institute of Standards and Technology) under the U.S. Department of Commerce, with experience spanning two administrations.

How much power the committee he's joining actually has

The Safety and Security Committee is chaired by Zico Kolter and exercises governance over safety practices across OpenAI's entire structure, including OpenAI Group PBC. According to OpenAI itself, this committee has the final word on frontier model launch decisions, as was the case with GPT-6 Astra. Christiano is also joining as a non-voting observer on the OpenAI Group PBC board.

How OpenAI is handling the conflict of interest

Since Christiano is keeping his position at CAISI — an agency that also oversees AI standards at the federal level — OpenAI announced he will recuse himself from any OpenAI-related matters within his government role, as well as from model evaluations within the safety committee. It's an unusual arrangement: one person with a foot in both the regulator and the safety governance board of the company being regulated, with explicit cross-recusal rules.