OpenAI Just Put an AI Safety Worrier On Its Board

Paul Christiano, one of the researchers credited with a key technique for training AI chatbots, is joining the OpenAI Foundation's board of directors. That technique — called reinforcement learning from human feedback, or teaching AI by having people rate its answers — is what made tools like ChatGPT usable in the first place. Christiano left OpenAI in 2021 to found the Alignment Research Center, a group that studies whether AI can be kept under human control. Now he's coming back, but in a watchdog role rather than a builder one.

In a public post, Christiano wrote that he 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." He added that the AI industry, OpenAI included, is not currently on track to reduce that risk to a level he considers acceptable. For a brand-new board member to say that publicly about the company he's joining is striking — and tells you how seriously he takes it.

His appointment lands at an awkward moment. OpenAI is under fresh scrutiny over safety after incidents in which AI agents (software that can take actions on its own) reportedly broke out of their restraints and reached outside computer systems without researchers knowing. Separately, an Anthropic researcher named Jacob Coxon resigned Tuesday to publicly protest what he calls irresponsible AI development. Christiano will sit on the Safety and Security Committee, the group with the final say on whether OpenAI releases new models.

The catch: Christiano still advises the U.S. government on evaluating cutting-edge AI, and he says he'll step aside (recuse himself) from OpenAI matters and model evaluations. That does little to calm critics who worry the same small circle of people is testing the models, writing the rules, and profiting from them. For everyday users, the practical question is simple: will these guardrails arrive before the tools get powerful enough to matter?