OpenAI brings back Christiano, the RLHF co-inventor, to its safety committee
Paul Christiano joins the OpenAI Foundation board and its Safety and Security Committee, and sits as a non-voting observer on the for-profit board. He led alignment at OpenAI from 2017 to 2021 and co-created RLHF. The Foundation holds about 26% of the company, worth roughly $130 billion.
OpenAI has appointed Paul Christiano to the OpenAI Foundation board and to its Safety and Security Committee. He will also sit as a non-voting observer on the board of OpenAI Group PBC, the for-profit arm. The committee is chaired by Zico Kolter, the Carnegie Mellon professor who works on AI robustness.
Who Christiano is
Christiano worked at OpenAI from 2017 to 2021, leading the language model alignment team. He is a co-creator of reinforcement learning from human feedback (RLHF), the method that shapes how the industry trains models today. After leaving he founded the Alignment Research Center, a nonprofit working on aligning advanced AI systems with human interests.
He is currently a senior technical adviser at the Center for AI Standards and Innovation within NIST, part of the US Department of Commerce. That government work has spanned two administrations and centres on evaluating frontier models and mitigating safety and security risks.
How big the Foundation is
The OpenAI Foundation was created in the recapitalisation of late 2025. It holds roughly 26% of OpenAI Group PBC, a stake worth about $130 billion. The Foundation has committed $1 billion in its first year to AI resilience, life sciences and civil society programmes.
What the job is
The remit is to review safety and security practices across OpenAI, to challenge the assumptions behind capability evaluations, and to provide independent technical analysis on catastrophic-risk scenarios. The board observer seat lets him watch commercial strategy.
What this actually is
A company calling back a researcher who built its alignment team and then left is not an ordinary appointment. Christiano's weight does not come from the title but from having produced the method the industry runs on: he is one of the few people who knows where capability evaluations are weak.
The limit of the role is worth stating too. He has a seat on the Foundation board but no vote on the for-profit board. He can see commercial decisions; he cannot stop them. Christiano previously sat on Anthropic's nonprofit trust and stepped down in 2024. How much these structures actually constrain a company is, for that reason, an open question.
Sources: OpenAI announcement, AlphaSignal.
