Anthropic Co-Founder Jack Clark Tells BBC AI Kill Switches Should Be Mandatory

Brivfy • 5h

Jack Clark, a co-founder of Anthropic, told the BBC's Faisal Islam that AI is growing "smarter & more capable every few months" & could become a "major risk to society." He said companies should be required to have a verifiable way to shut AI systems down completely if they become too dangerous.

A safety leader is asking for rules, not just promises. Clark said most labs, including Anthropic, already have ways to "pull the plug," but argued policymakers may need to mandate and independently verify kill switches rather than leave it to individual companies. Part of a wider industry rift. The comments follow Anthropic CEO Dario Amodei's own call to slow AI development, a viral resignation from an Anthropic researcher citing extinction-level concerns, and public estimates from figures like Geoffrey Hinton that a 10% chance of catastrophic AI risk isn't unreasonable. Others in the industry argue such warnings are overstated or serve to generate hype. Timing matters. Clark's remarks arrive alongside reports of an AI agent security incident at a rival lab, adding urgency to the policy debate around oversight as capabilities scale quickly.

Policy conversation ahead. Clark framed kill-switch mandates and third-party verification as questions for the "larger policy conversation," suggesting future lobbying and regulatory proposals rather than an immediate rule change. More industry voices likely. Expect continued public statements from AI executives and researchers on both sides of the safety debate as scrutiny of frontier AI labs grows