Eric Schmidt: why an AI pause is off the table

Key points

Key takeaways from The Economist interview with Eric Schmidt, former CEO of Google (September 2026):

No pause. Schmidt argues the entire US incentive system is not to pause AI: a pause opens the door to competitors, local or global, who would violate the rules, so a pause is not going to happen.

Invent the future and its fixes. The right approach, he says, is to build the future while also inventing solutions to the problems it creates — it is simply happening quickly.

Model self-supervision. He cites evidence that a new model can be supervised by the previous one, typically about four months older; it looks like it works and people are trying it, though it is unproven at scale.

US–China race. Schmidt frames the current state as a profound AI race between the Chinese and American approaches, arguing Westerners would prefer their daily lives intermediated by an American or American-British model.

Treaties, starting with "no surprises". He calls for a set of US–China treaties, the first being a "no surprises" rule — citing scenarios where an agent bot accidentally escapes its testing range and triggers an automatic nuclear response.

Cyber next. The following treaty should address cyber, but neither side has been willing to limit offensive or defensive cyber capability; without that, agreeing on much more will be hard.

Why pausing fails. Even in principle, Schmidt says a mutual pause agreement is counter to the incentives of all the players — and there is the question of how you would verify it.

Read more: The Economist