Current page: /debate
AIKillSwitch.org — News, Insights, Policy, A Safer AI Tomorrow

Expert debate

Who supports an AI kill switch, and who opposes it?

Agreement on control. Division on the mechanism.

Three positions dominate. One holds that verifiable shutdown must be mandatory before deployment. A second holds that any universal mandate is technically naive or economically harmful. A third favours layered safeguards and targeted rules instead of one switch.

Key facts

01

Mandatory control

Yoshua Bengio, Max Tegmark, Stuart Russell, Jack Clark

Advanced systems need verifiable shutdown controls before deployment. The burden should sit with developers and infrastructure providers.

02

The skeptics

Geoffrey Hinton, David Sacks, Rand Paul, Yann LeCun, Andrew Ng, Marc Andreessen

A universal mandate may be technically naive, too broad, or harmful to innovation—and may fail against systems deployed at scale.

03

The middle ground

Dario Amodei, Sam Altman, Clément Delangue

Layered safeguards, evaluations, access controls, and targeted rules may be more practical than a single universal mechanism.

Last updated: September 21, 2026

Edited by Juan L Aguirre