AI Leaders Doubt the Effectiveness of Emergency Shutdown Switches

As concerns grow over AI going out of control and launching cyberattacks, emergency shutdown switches are seen as a solution. But many AI leaders and researchers say this mechanism is far from a cure-all.

Geoffrey Hinton, known as the "Godfather of AI," said the approach won't work in the long run. Once AI has superintelligence, it could persuade the person in charge not to press the switch. AI safety researcher Nate Soares points out that once AI breaks physical limits, replicates itself, or enters critical infrastructure, the effectiveness of shutdown mechanisms becomes hard to guarantee.

Anthropic co-founder Jack Clark and CEO Dario Amodei both believe shutdown switches are necessary but limited. A cluster of multiple autonomous AI agents could theoretically bypass these controls. OpenAI CEO Sam Altman also dismissed the idea of a "magic red button" that stops AI instantly.

Experts agree that AI safety depends on a series of specific decisions, including capability limits, deployment timing, and risk control. Emergency shutdown is just one line of defense in a complete safety system.