OpenAI’s Safety Board Adds a Longtime Skeptic of the Industry’s Own Safeguards

OpenAI announced on September 9 that Paul Christiano, the researcher whose foundational work on reinforcement learning from human feedback (RLHF) helped shape how today’s chatbots are trained, and who has spent years publicly questioning whether the AI industry’s safety measures are adequate, is joining the OpenAI Foundation Board. He will also sit on the Foundation’s Safety and Security Committee (SSC), the board committee that governs safety and security practices across all of OpenAI.

“AI capabilities have advanced very rapidly in the last year and alignment remains a difficult technical problem, making the Safety and Security Committee’s responsibility more important and more challenging than ever,” Christiano said in OpenAI’s announcement. “I am excited to join the team to help.”

The appointment is notable less for who Christiano is, he is well known in AI safety research circles, than for what OpenAI itself says about why it wants him. The company’s own announcement describes him as someone who “has been an independent voice on whether the industry’s safeguards are adequate,” a line that amounts to OpenAI publicly acknowledging it is recruiting one of the field’s more persistent skeptics into a seat with real institutional standing.

A Board Seat With an Asterisk

Christiano’s new role actually spans two different bodies with very different amounts of power attached. On the board of OpenAI Group PBC, the for-profit public benefit corporation that runs OpenAI’s commercial operations, he will serve only as a non-voting observer. On the OpenAI Foundation Board, the nonprofit that controls the for-profit Group and holds a significant equity stake in it, he becomes a full independent director. OpenAI’s own structure page now lists him alongside Foundation chair Bret Taylor, Adam D’Angelo, Sue Desmond-Hellmann, Zico Kolter, retired General Paul Nakasone, Adebayo Ogunlesi, Nicole Seligman, David Vélez, and Robin Vince, plus CEO Sam Altman.

Taylor framed the appointment in similar terms to Christiano’s own statement. “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,” he said. “His experience working on frontier AI safety and standards in government, together with his technical judgment, will strengthen the Board’s oversight and the Safety and Security Committee’s work as OpenAI continues to develop and deploy frontier AI.”

His most consequential seat is the one on the SSC, the Foundation Board committee that Kolter chairs. The SSC’s job is governance over safety and security practices across the whole organization, including the Group PBC, and its chair has previously described its authority in concrete terms: OpenAI can “request delays of model releases until certain mitigations are met.” Christiano joins Kolter, D’Angelo, Nakasone, and Seligman on that committee, per OpenAI’s own account of its formation and outside reporting on its 2024 reconstitution as an independent body.

One footnote in OpenAI’s announcement limits how directly Christiano can use that seat day to day. As a sitting Senior Technical Advisor at the U.S. government’s Center for AI Standards and Innovation (CAISI), he will recuse himself “from all OpenAI-related matters as well as all model evaluations.” That leaves an open question about how much hands-on oversight the seat actually confers, versus the credibility his name lends to the committee simply by being on it.

Where Christiano’s Skepticism Comes From

Christiano led OpenAI’s language model alignment team from 2017 to 2021 and, during that stretch, contributed foundational work on RLHF: using human preference rankings to steer a model’s behavior, a technique that now underlies most commercial chatbot training. He left in 2021 to found the Alignment Research Center (ARC), a nonprofit built around a harder theoretical problem than RLHF ever tried to solve: verifying whether an advanced AI system’s stated reasoning actually matches what it is doing internally.

Two years after leaving, Christiano put a number on his concern that stuck, in an interview on the Bankless podcast that Fortune reported on in early May 2023. “Overall, maybe you’re getting more up to a 50-50 chance of doom shortly after you have AI systems that are human-level,” he said, and Fortune’s own headline on the interview put the odds of an outright AI takeover leaving “many or most humans dead” at “10 to 20%.” Those were not casual remarks from an outsider. They came from someone who had spent years building the exact training technique the industry now uses to make its models behave, before deciding it was not enough on its own.

Since leaving OpenAI, Christiano has also taken government roles that put him in a position to evaluate frontier models from outside any single lab: first at the U.S. AI Safety Institute, then at its successor, CAISI, assessing capabilities with national security implications across the industry. OpenAI’s own announcement credits that “government experience spanning two administrations” as part of what he now brings to the Foundation Board.

A Governance Structure Built Under Legal Pressure

Christiano’s seat did not appear in a vacuum. It sits inside a governance structure OpenAI rebuilt through a recapitalization it announced on October 28, 2025, splitting the organization into the nonprofit OpenAI Foundation and the newly created OpenAI Group PBC. Under that structure, the Foundation holds conventional equity in the Group and retains special voting rights that let it appoint or replace every member of the Group’s board of directors at any time.

OpenAI’s own announcement of the Christiano appointment is explicit that the restructuring followed roughly a year of engagement with the California and Delaware Attorneys General, whose reviews of the nonprofit-to-for-profit conversion “helped strengthen the governance framework.” The SSC itself predates that process. OpenAI’s board formed it in May 2024 as a recommendation-only body, then reconstituted it in September 2024 as an independent oversight committee chaired by Kolter, with D’Angelo, Nakasone, and Seligman as its other members, according to reporting on the change at the time (see this Associated Press account). Kolter now serves exclusively on the Foundation board rather than also sitting on the Group PBC board, a separation OpenAI says is meant to preserve the committee’s independence from the commercial side of the business.

Why It Matters Now

The appointment lands in the middle of a stretch in which OpenAI’s own safety commitments have been repeatedly tested in public. The company disbanded its dedicated Preparedness team earlier this year amid a wave of safety-focused departures, then disclosed a new round of technical safeguards after briefly pausing training on its Astra model over flagged cyber capabilities. It also follows an August report card from the AI-safety nonprofit Guidelight that found labs across the industry still struggle to document how they would contain a model that started behaving in unintended ways, a question the SSC exists specifically to be able to ask.

Adding a director whose public track record is built on skepticism toward exactly that kind of self-assessment can be read two ways. It could be a genuine attempt to put a credible outside check inside a body that already has real standing to slow OpenAI down. Or it could be a way to associate the company’s governance with a well-regarded critic without materially changing what that critic can act on, given his recusal obligations. OpenAI’s own framing, that it wants “people prepared to challenge prevailing assumptions” on its board, suggests the company is betting reputationally that the first reading is the one that sticks.