Meta’s Oversight Board Urges AI Companies to Give Safety Committees Real Power

Meta’s Oversight Board wants AI companies to give their promised safety watchdogs enough power to challenge management. In interviews published by NBC News on October 7, members warned that committees without independence and authority risk becoming window dressing. Their warning accompanies an open letter with six recommendations responding to the White House Accord on Super Intelligence.

The September 29 accord brought together Google, Anthropic, Meta, OpenAI, xAI and Nvidia. It calls for four layers of safety checks: internal controls, an internal team checking those controls, independent outside assessments and oversight by an independent committee of each company’s board of directors. The agreement is voluntary, not a legally binding requirement.

Under that structure, the board committee would receive reports from the teams running safety controls and from internal and external auditors or evaluators. It would also ensure that identified problems are addressed. That places the committee within the company’s governing board, rather than creating a separate outside body.

The Oversight Board’s letter calls for a concrete mandate, structural stability and independence, expertise from different sectors, and transparency. Members’ interviews sharpened those principles into practical tests. Nighat Dad, a board member who leads Pakistan’s Digital Rights Foundation, emphasized three questions about how an oversight body operates:

• Can it access nonpublic company information, including material that might reflect badly on the business?
• Is it separate from the company, with control over its own budget and membership?
• Are its decisions binding, giving the company an obligation to act rather than merely consider its advice?

If it’s going to be genuine oversight, there’s going to be friction between the company and the body

Paolo Carozza, Oversight Board member and University of Notre Dame professor, speaking to NBC News

Meta’s Oversight Board is legally independent from the company and does not include Meta directors. Its members include journalists and human rights legal scholars. It has final authority over certain content-moderation decisions, but its advice on other topics is nonbinding.

Its financial separation also has limits. Meta has supplied more than $300 million since 2019, held in an independent trust. The company controlled part of the board’s membership selection process until 2023. Those arrangements form part of the experience members are drawing on as AI companies design their committees.

Stanford law professor Evelyn Douek, who has studied the board, called it more of a cautionary tale than a promising model. She pointed to its narrow content-moderation powers, rather than binding authority over Meta’s wider policy and safety decisions. Douek also questioned whether a body reliant on company funding can meaningfully challenge its sponsor.

Carozza agreed that a different design could have made the board more effective. He said members were offering lessons from six years of experience, not presenting themselves as the solution. Their intervention comes before AI companies cement their oversight plans.

The committees’ practical influence remains unsettled. Douek said it was too early to judge their weight, with many mechanisms still unclear. Beyond formal authority, she raised another constraint: whether oversight bodies will have enough technical expertise and time to investigate complex AI safety problems.