Facebook Co-Founder Proposes an "Office of AI Supervision" Inside AI Labs

Chris Hughes, a co-founder of Facebook, has called for AI companies to be supervised the way banks are. In an op-ed published by the Financial Times on 23 September, he proposes creating an "Office of AI Supervision" that would embed government supervisory teams inside frontier AI laboratories, with supervisors serving three-year terms to limit regulatory capture, according to the FT piece as summarized by Techmeme.

What the proposal calls for

Hughes, who left Facebook in 2007, is the author of "Marketcrafters: The 100-Year Struggle to Shape the American Economy" and a non-resident fellow at the Brookings Institution. He argues that the two industries resemble each other: both change rapidly and reach throughout the economy, so officials have to weigh the benefits of innovation against the risks it creates. Rather than auditing labs from outside, his office would put public supervisors inside the companies, on fixed three-year terms designed to keep them from growing too close to the firms they oversee.

The argument Hughes takes up

The op-ed engages with Dario Amodei, Anthropic's chief executive, who argued in his September essay "We Must Pace the Frontier" that independent evaluators should have desks inside AI companies. Amodei said Anthropic would unilaterally give outside reviewers access comparable to its internal risk teams — desks, badges and company laptops — plus the right to publish findings without the company's editorial control, subject to narrow redactions. He compared the arrangement to bank supervision, where regulators are sometimes embedded alongside employees, and called on governments to require other frontier labs to match. OpenAI's Sam Altman said his company would do the same.

Why it matters

Banking supervisors can order an institution to stop a practice, restrict its growth or, in extreme cases, close it. The evaluators proposed for AI labs so far would be able to investigate and report, but not to halt a model's training or release — a gap that critics of voluntary oversight have highlighted. Hughes's proposal goes further by placing government teams, rather than outside contractors, inside the labs, and by limiting their time in post. The op-ed lands as governments argue over the same question: in his address to the UN on 22 September, US President Donald Trump rejected an international pact on AI risks and said he wanted artificial intelligence renamed "super intelligence", according to coverage tracked by Techmeme.