Breaking

Anthropic Co-Founder Jack Clark Says Mandatory AI Kill Switches May Be Needed

September 16, 2026 at 11:08 EDT

Policy & Regulation
Security
AI Agents

Anthropic co-founder and policy chief Jack Clark said governments may eventually need to require frontier AI companies to maintain independently verifiable ways to shut down dangerous systems, escalating a fast-moving debate over how to control autonomous AI agents. The comments put a concrete regulatory demand around a broader industry push for stronger outside oversight of the most capable models.

Sep. 2026 · Anthropic

AI shutdown controls move into policy

Jack Clark says lawmakers may eventually require frontier AI companies to maintain kill switches that outside evaluators can verify.

Anthropic founders

days to report qualifying incidents under the bill

$100M

compute threshold in the proposed bill

Emergency fines dwarf general penalties

$2M/daygeneral violations

$20M/dayemergency order defiance

Each block represents $2 million per day.

What a mandatory control regime would ask

Stop inference

Pause model outputs when risks are detected.

Cut access

Terminate user or account access.

Verify controls

Let outside evaluators check that safeguards work.

Supportive view

Verifiable shutdown capability is seen as a minimum control layer for autonomous agents that can act across software environments.

Critical view

Critics warn that a kill switch can be vague, oversimplified or vulnerable if attackers target the control mechanism itself.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.