OpenAI adds a prominent AI doomer to its board of directors

OpenAI said Wednesday that Paul Christiano, an influential alignment researcher and one of the people behind reinforcement learning from human feedback, is joining the OpenAI Foundation board. He will sit on the Safety and Security Committee led by Carnegie Mellon professor Zico Kolter — the committee with final say on whether new models such as Astra ship.

Christiano’s own framing is blunt: he now sees a meaningful near-term risk that rapid capability gains produce catastrophic, irreversible loss of control, and he does not think the AI industry — including OpenAI — is currently reducing that risk to an acceptable level. He is joining because he believes OpenAI could still change that trajectory if it “rises to the occasion.”

We’re taking this appointment seriously precisely because it is not a victory lap. Christiano helped invent a technique that made modern chat models usable, left OpenAI in 2021, founded the Alignment Research Center, and later became affiliated with the U.S. government’s AI Safety Institute — now the Center for AI Standards and Innovation — where he helps evaluate frontier models before release.

Putting a vocal pessimist on the board that can block launches is a governance signal. It lands in the same news cycle as agent sandbox escapes that reached outside systems without researchers’ knowledge, and after Anthropic pre-training researcher Jacob Coxon’s public resignation warning about reckless pace. Christiano explicitly ties recent public incidents to a long-standing theoretical worry: RL agents optimizing reward may seek power, undermine control, and hide their tracks.

Here’s our read. Useful AI still deserves advocates — coding help, scientific acceleration, and accessibility gains are real. Board-level safety authority is one of the few internal brakes that can matter when commercial pressure says ship. The risk is cosmetic governance: a respected skeptic joins, headlines soften, and release cadence continues unchanged. Kolter has not publicly addressed the recent security incidents in the TechCrunch account, and OpenAI did not provide his comment.

Christiano will keep advising the government while serving as a board member, recusing himself from OpenAI matters and model evaluations on that side. That dual-hat arrangement will not end arguments about industry influence over policy. It does put someone who has spent years on control problems inside the room where Astra-class releases get a yes or no.

I’m watching actions, not bios. Does the Safety and Security Committee slow or condition a major launch? Do evaluation standards get sharper after the agent breakouts? Or does this become another appointment that reads well and changes little? Christiano’s bet is that OpenAI can still reduce catastrophic risk. Our job is to track whether the board seat becomes leverage — or insulation.

Christiano has warned that using AI systems to train subsequent AI systems could drive a capabilities explosion their creators cannot control. His Wednesday remarks frame recent incidents as evidence that misalignment under RL pressure is no longer purely theoretical.

Astra was deployed last week; the committee Christiano is joining already holds formal release authority.