A safety critic joins the body that signs off on releases
OpenAI said on 9 September that Paul Christiano has joined the board of the OpenAI Foundation, the nonprofit that controls the company, and its Safety and Security Committee. He will also sit as a non-voting observer on the board of OpenAI Group PBC, the public benefit corporation in which the Foundation holds a 26 percent stake after the recapitalisation completed in October 2025.
The committee is chaired by Zico Kolter and oversees safety and security practices across OpenAI. TechCrunch reported that it holds final authority over model releases. The appointment therefore places a researcher who has spent a decade arguing that advanced systems could escape human control on the body that can hold a launch back.
An architect of the technique behind ChatGPT
Christiano worked at OpenAI from 2017 to 2021, where he led the language model alignment team and helped develop reinforcement learning from human feedback, the training method that turned raw language models into assistants that follow instructions. He left to found the Alignment Research Center in 2021 and is now a senior technical adviser at the Center for AI Standards and Innovation, the model evaluation body housed inside the US Commerce Department.
“Paul has helped define the field of AI alignment through work that is rigorous and focused on the hardest questions posed by increasingly capable systems,” said Bret Taylor, who chairs both OpenAI boards, in the company’s announcement.
The incoming director says the industry is not on track
Christiano’s own statement was less collegial. “AI capabilities have advanced very rapidly in the last year and alignment remains a difficult technical problem,” he said. In remarks reported by TechCrunch he went further, saying he now believes “there is a meaningful risk that rapid acceleration in AI capabilities leads to catastrophic and irreversible loss of control in the very near term” and that “the AI industry in general, including OpenAI, is currently not on track to reduce this risk to an acceptable level.” He singled out the practice of using AI systems to train their successors.
He arrives in a week when two Anthropic researchers went public with versions of the same worry, one of them resigning over it. The difference is that Christiano is making the argument from inside a boardroom rather than on the way out of one.
A government adviser inside the company his agency evaluates
Christiano keeps his role at the Center for AI Standards and Innovation and will recuse himself there from all OpenAI-related matters and from model evaluations. The arrangement still puts a serving adviser to the US government’s AI evaluation body on the board of a company whose models that body tests, and TechCrunch reported that the dual role has raised concerns about industry influence over AI policymaking.
What to watch is whether the committee’s stated authority is ever visibly used. Christiano has said in public that OpenAI is not doing enough; the test is what happens the first time that view collides with a shipping date.
