Paul Christiano Joins OpenAI Foundation Board and Safety Committee

OpenAI appointed alignment researcher Paul Christiano to the OpenAI Foundation Board on September 9, 2026, placing him on the nonprofit body that controls the company’s commercial public benefit corporation. He will also join the Foundation’s Safety and Security Committee, which has formal authority to require mitigations and halt releases of OpenAI models or AI systems.

Christiano will not become a voting director of OpenAI Group PBC. Instead, he will attend its board as a non-voting observer. That distinction matters because it gives him access to the commercial company’s board-level discussions while locating his voting authority within the controlling nonprofit.

The appointment brings a former OpenAI alignment lead and U.S. government frontier-model evaluator into one of the company’s most consequential oversight positions. It does not itself change any OpenAI model, product, release schedule, or safety threshold.

1. Christiano Takes Three Connected Governance Roles

Christiano is joining the OpenAI Foundation Board as a director, the Foundation Board’s Safety and Security Committee as a member, and the OpenAI Group PBC Board as a non-voting observer.

The Foundation is not merely OpenAI’s philanthropic arm. Under the structure completed in October 2025, it controls OpenAI Group PBC through special voting and governance rights. The Foundation appoints every director of the Group board and can replace those directors at any time.

OpenAI’s current governance page lists 11 Foundation directors: ten whom the company describes as independent, including Christiano, plus CEO Sam Altman. Bret Taylor chairs both the Foundation and Group boards.

Most Foundation directors have also served as directors of OpenAI Group. Safety and Security Committee chair Zico Kolter was the initial exception, serving only as a Foundation director and attending Group board meetings as a non-voting observer. Christiano’s appointment creates another split role of this kind.

The structure gives Christiano a vote on the controlling nonprofit board but no vote in his observer capacity at the commercial company. An observer can participate in board-level information flows and deliberations, but the appointment announcement does not assign Christiano an individual veto over Group decisions.

His committee membership is more consequential than an ordinary advisory position because the Safety and Security Committee exercises authority delegated through OpenAI’s corporate governance arrangements. Christiano will work alongside Kolter, whom OpenAI appointed to chair the committee in 2024.

2. The Safety Committee Can Intervene in Model Releases

OpenAI established the Safety and Security Committee in 2024 and initially tasked it with reviewing safety and security processes. After a 90-day review, the company converted it into an independent board oversight committee responsible for safety and security across model development and deployment.

OpenAI said the committee would receive briefings on evaluations for major model releases, review technical assessments and post-release monitoring, and participate with the full board in launch oversight. The company also said the committee and board could delay a release until identified safety concerns were addressed.

Those powers were subsequently recorded more explicitly during OpenAI’s 2025 recapitalization. A memorandum signed by OpenAI and the California attorney general states that the committee remains part of the nonprofit rather than the PBC and has an effective approval right over Group actions involving safety and security.

The memorandum says the committee may require mitigation measures “up to and including halting the release of models or AI systems,” even when OpenAI’s applicable risk thresholds would otherwise permit release. OpenAI must give the California attorney general advance notice of any material change to that authority.

The agreement also requires the PBC board to consider only OpenAI’s mission—not shareholders’ financial interests or other interests—when addressing enterprise safety and security matters. The Foundation’s control of Group board appointments provides an additional route for nonprofit oversight beyond the committee itself.

Christiano therefore joins a body with documented powers to demand safeguards and stop releases. Whether and how those powers will be exercised remains a separate question: neither OpenAI nor Christiano announced a pending intervention, revised safety policy, or model-specific review with the appointment.

3. His Background Combines Alignment Research and Government Evaluations

Christiano worked at OpenAI from 2017 to 2021, where he led alignment research and contributed foundational work on reinforcement learning from human feedback. RLHF uses human judgments to train a reward model and then optimize an AI system toward behavior preferred by evaluators; variants of the approach became central to training conversational AI systems.

After leaving OpenAI, he founded the nonprofit Alignment Research Center. He also launched an initiative for independent evaluations of frontier models that later became Model Evaluation and Threat Research, or METR.

The U.S. Department of Commerce appointed Christiano head of AI safety at the U.S. AI Safety Institute in April 2024. The institute, housed within the National Institute of Standards and Technology, tasked him with designing and conducting frontier-model evaluations, particularly tests for capabilities with national-security implications, and contributing to associated risk mitigations.

That institute was later succeeded by NIST’s Center for AI Standards and Innovation. OpenAI identifies Christiano as a senior technical adviser at CAISI and says his government experience spans two presidential administrations.

His continuing government role creates an evident conflict-management requirement. OpenAI’s announcement says that while serving as a senior technical adviser, Christiano will recuse himself from every government matter related to OpenAI and from all model evaluations. The disclosed restriction is broad: it is not limited to evaluating OpenAI models.

The announcement does not say he is leaving ARC or his government advisory position. It also does not disclose compensation, the term of his Foundation appointment, or additional committee-specific recusals.

4. Christiano Says the Appointment Is Not an Endorsement

In a personal statement published the same day, Christiano described loss-of-control risk from advanced AI as acute. He argued that automating AI research could create a feedback loop in which improved AI systems accelerate the development of still more capable systems.

He gave subjective, all-things-considered risk estimates of 4% over the next year and 15% over the next three years for catastrophic and irreversible loss of control. Christiano emphasized that these figures communicate his personal uncertainty rather than results generated by a precise or stable forecasting model.

He also said the AI industry, including OpenAI, was not currently on track to reduce the risk to an acceptable level. That statement goes beyond OpenAI’s announcement, which describes him more generally as someone willing to challenge assumptions and question whether industry safeguards are adequate.

Christiano explicitly cautioned against treating his decision to join as either an endorsement or criticism of OpenAI’s existing safety practices. He said developers should be judged by externally verifiable conduct and results, identifying stronger mitigations, transparent evidence about risks and safeguards, and shared safety standards as practical measures.

The appointment’s immediate effect is institutional rather than technical. OpenAI has added Christiano’s judgment to the board that controls the Group and to the committee empowered to examine evaluations, require mitigations, and block releases. Evidence of impact will depend on subsequent observable actions, such as changes to published safeguards, evaluation disclosures, release decisions, or the committee’s use of its intervention authority.

Frequently Asked Questions

Is Paul Christiano joining OpenAI as an employee?

No. OpenAI announced him as a Foundation director, Safety and Security Committee member, and non-voting observer on the Group board. It did not announce an operating or research job.

Can Christiano personally stop an OpenAI model release?

OpenAI has not granted him an individual veto. The Safety and Security Committee collectively has documented authority to require mitigations, including halting a release.

Does he have a vote on the OpenAI Group PBC Board?

No. Christiano is a non-voting observer on the Group board. He is a voting director of the controlling OpenAI Foundation.

Will he continue working with the U.S. government?

OpenAI says he remains a senior technical adviser at NIST’s CAISI. He must recuse himself from all OpenAI-related government matters and all model evaluations.

Does the appointment change OpenAI’s products or safety rules?

No product or policy change was announced. The appointment changes the membership of OpenAI’s governing Foundation board and its Safety and Security Committee.

Sources

• Original OpenAI announcement on X
• OpenAI: Paul Christiano joins OpenAI Foundation Board
• Paul Christiano: Personal statement on joining the OpenAI board
• OpenAI: Our structure
• OpenAI: An update on our safety and security practices
• California Attorney General: Memorandum of Understanding with OpenAI
• NIST: Paul Christiano
• NIST: U.S. AI Safety Institute leadership expansion