Meta is the only major US AI company that has not agreed to let the federal government review its most capable models.
The Trump administration has been pushing Meta to submit its frontier AI models for federal security review, according to a New York Times report. The pressure has come through direct emails from Washington. Every other major US AI developer has already agreed to the reviews. Meta has not.
That gap matters because Meta's models — including its Llama family — are open-source and widely deployed, which means the security surface is broader than a typical closed model. If the government's concern is about what frontier AI can do in the wrong hands, an open-source lab that hasn't signed on to review is a meaningful blind spot in any oversight framework.
Meta has positioned its open approach as a public good, and perhaps it is — but "we publish the weights for everyone" is also a convenient argument against the kind of closed-door security review that its competitors have quietly accepted.