Demis Hassabis urges U.S.-led agency to vet new AI models

Technology Jul 17, 2026

Demis Hassabis, who helped start DeepMind, is calling for a fresh approach to AI regulation.

He suggests creating a U.S.-led agency (kind of like FINRA in finance) that would check new AI models for safety and risks before they go public.

Hassabis feels this is urgent, warning that unchecked AI could bring real dangers.

Fast oversight for frontier labs

This agency wouldn't be your usual slow government setup; it'd be faster and more flexible.

The main target? "Frontier labs" or teams building the most advanced AI.

Only models with serious capabilities, like hacking skills or following tricky instructions, would need to pass these checks.

Hassabis wants practical oversight to keep reckless development in check and avoid endless debates over academic or open-source projects.