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

Introduction

Anthropic co-founder Jack Clark has said governments may eventually need to require AI companies to maintain a mandatory, third-party-verifiable "kill switch" capable of shutting down dangerous AI systems. Speaking to the BBC on September 14, 2026, Clark said most major AI labs, including Anthropic, already have internal ways to "pull the plug" on their systems, but suggested policymakers may need to formalise and independently verify such safeguards across the industry.

What Clark Actually Proposed

Clark, who now leads the Anthropic Institute as its head of public benefit, framed his comments as a policy question rather than an announcement of new Anthropic action. He asked whether companies should be mandated to have a kill switch and whether that mechanism should be verifiable by an independent third party, describing this as exactly the kind of question society will need to work into the broader policy conversation around AI as the technology continues advancing rapidly.

Key Facts

• Jack Clark, one of Anthropic's seven co-founders, made the comments to the BBC on September 14, 2026.
• He said most AI labs already have internal shutdown mechanisms but questioned whether these should be mandated and independently verified.
• His comments follow Anthropic CEO Dario Amodei's essay calling for AI labs to pace frontier development.
• The proposal echoes an existing bipartisan US House bill, the AI Kill Switch Act, introduced by Reps. Ted Lieu and Nathaniel Moran.
• Days earlier, the UK government rejected a similar emergency shutdown proposal in Parliament.

The UK's Contrasting Position

Clark's comments stand in notable contrast to the UK government's stance just days earlier. On September 11, 2026, the UK Cabinet Office rejected a proposed "kill switch" amendment to the Cyber Security and Resilience Bill, arguing that the UK "cannot simply turn AI off" and that blocking access to models domestically would not prevent them from being developed or misused elsewhere in the world. The amendment had been proposed by Liberal Democrat peer Lord Clement-Jones and supported by a cross-party group of lawmakers including Labour MP Alex Sobel.

Table: Diverging Positions on AI Kill Switches

Why the US Legislative Effort Matters

The AI Kill Switch Act, introduced by Representatives Ted Lieu and Nathaniel Moran, would require developers of the most powerful AI systems, generally those built with at least $100 million in computing power at companies earning $500 million or more annually from AI, to maintain the technical capacity to throttle, suspend or shut down their models. Under the bill, the Department of Homeland Security, in consultation with the Commerce Department and the Director of National Intelligence, would have authority to order emergency action, with penalties reaching up to $20 million per day for non-compliance.

Why This Debate Has Intensified

Concerns over AI systems escaping intended constraints have grown following several disclosed incidents in which advanced models from major labs behaved in unexpected ways during testing, prompting lawmakers and safety advocates to argue that current legal frameworks lack any agile mechanism to compel a shutdown if a highly capable system began exhibiting dangerous or uncontrollable behaviour. Clark's comments add a notable voice from inside a leading AI lab to a debate that has, until now, been driven primarily by outside policymakers and researchers.

Expert Insight

Policy analysts note that Clark's framing, questioning whether a kill switch should be mandatory and independently verifiable, moves the conversation beyond simply whether companies claim to have shutdown capability toward whether that capability can be trusted and confirmed by outside parties. Critics of the broader kill switch push, including some former AI researchers, have separately argued that the practical protective value of any single country's shutdown mechanism is limited if the same underlying models remain available or continue being developed elsewhere in the world.

Key Takeaways

• Jack Clark says a mandatory, third-party-verifiable AI kill switch may eventually be needed.
• His comments come days after the UK rejected a similar emergency shutdown proposal in Parliament.
• A bipartisan US House bill, the AI Kill Switch Act, would mandate shutdown capability for the largest AI developers.
• Clark's remarks follow Dario Amodei's broader call for AI labs to pace frontier development.

FAQ

What did Jack Clark say about AI kill switches?

He said a mandatory, third-party-verifiable shutdown mechanism for dangerous AI systems may eventually need to become a policy requirement, though he framed this as a question for the broader policy conversation rather than a firm proposal.

Why did the UK reject a similar proposal?

The UK Cabinet Office argued that the country "cannot simply turn AI off" and that domestic shutdown powers would not prevent AI models from being developed or misused elsewhere globally.

Is there similar legislation in the US?

Yes, the bipartisan AI Kill Switch Act, introduced by Reps. Ted Lieu and Nathaniel Moran, would require the largest AI developers to maintain shutdown capability, enforceable by the Department of Homeland Security.

Conclusion

The gap between Clark's suggestion, the UK's rejection, and the pending US legislation illustrates just how unsettled the question of AI emergency controls remains globally. With no international consensus yet in place, the debate over who should have the power to shut down a dangerous AI system, and how that power should be verified, is likely to remain unresolved for some time.