The Off Switch

Who, today, actually has the power to stop a dangerous AI model?

In June 2026, a cabinet secretary caused two frontier AI models to be switched off for 19 days, under an authority the government never publicly explained. Months later, Anthropic and OpenAI were negotiating in public with the governor of California over the terms on which a state could stop them. The answer to the question of who holds the switch turns out to be unsettling: many people are partly in charge, and nobody is clearly in charge.

The Off Switch follows the fight over the rules for powerful machines. It walks through what it actually means to stop a model, the warning shots that preceded the debate, the promises labs make to themselves, and the boards, trusts and open letters that stand in for law. It covers the state bills that would build a brake, the preemption fight in Washington, the politics and money of the pause, a contractor who said no to the Pentagon, the liability theories heading for the courtroom, the question of whether a model is speech, and the lever Brussels holds over everyone else.

The book sets out the strongest arguments on every side, from safety advocates to industry and civil libertarians, and does not tell you how to vote. It gives you the facts, the statutes, the procedures and the odds that a shutdown decision would actually hold.

If you want to understand AI regulation, the kill switch debate and who is really in control of one of the most powerful technologies of the decade, this is the book to read first.