A Brighton-born pioneer in artificial intelligence has warned that governments may need to introduce mandatory “kill switches” to prevent powerful AI models from causing global harm.
Jack Clark, 37, the co-founder of San Francisco-based Anthropic, compared the current state of the industry to a car that has been built with an accelerator but no way to stop.
The 37-year-old, who attended Varndean School in Brighton, suggested on Monday that third-party verifiable mechanisms to shut down AI systems should be made a legal requirement if the technology shows signs of “rogue” behaviour.
Industry warning vs government resistance
The call for stricter controls comes as a direct challenge to the UK government’s current stance. During the week ending 13 September 2026, ministers reportedly rejected a specific proposal for mandatory “kill switches”, preferring to maintain a “principles-based” regulatory approach rather than passing new statutory laws.
Mr Clark’s intervention highlights a growing divide between the world’s leading AI laboratories and political leaders. While tech companies have historically lobbied against regulation, both Anthropic and its rival OpenAI are now urging lawmakers to act.
On 14 September 2026, OpenAI’s head of policy in Europe, Tom Duff Gordon, also called on UK legislators to capitalise on a “political window” to bring the industry under tighter legal oversight.
What is an AI ‘kill switch’?
The proposed “kill switch” is not a physical button but a set of protocols and technical overrides. Under Mr Clark’s proposal, these would be:
- Mandatory: Required by law for any company developing “frontier” models.
- Third-party verifiable: Independent experts or government bodies must be able to prove the switch works without needing to see a company’s trade secrets.
- Emergency-focused: Designed to be used only if a model begins to demonstrate uncontrollable or dangerous capabilities.
“We are currently in a race where everyone is pressing the gas pedal,” Mr Clark said during an interview on Monday. “We need to ensure that as we build these increasingly powerful systems, we also build the brake pedal.”
Global security concerns
The debate coincides with a shift in how the UK monitors the technology. Tension is mounting as Anthropic prepares for an initial public offering (IPO) on the NASDAQ exchange, targeted for October 2026.
While UK ministers have so far resisted mandatory shutdown powers, the government continues to weigh the risks of stifling innovation against the safety warnings issued by industry experts, according to Computing UK.