A co-founder of Anthropic has called for companies developing artificial intelligence to be legally required to provide a mechanism capable of shutting down AI systems, as concerns grow over the potential risks posed by rapidly advancing technology.
Jack Clark told the BBC that a “kill switch”, subject to verification by an independent body, should be made mandatory for AI companies. He said most AI research laboratories already have ways to “pull the plug” on their systems, including Anthropic, but argued that lawmakers may need to require such safeguards formally.
Clark also said the question of whether such a shutdown mechanism could be activated by third parties should be considered. His comments add a new dimension to the debate over how AI systems should be controlled and who should ultimately have the authority to intervene if serious risks emerge.
The discussion comes as the US political leadership considers its approach to AI regulation. House Speaker Mike Johnson said on Monday that he expected a meeting between US President Donald Trump and representatives of the AI industry to take place in the coming days, although no meeting had yet been scheduled.
Johnson said the meeting would involve an open discussion about the responsibility of technology companies for AI safety and the possible role of government. He expressed concern about measures that could slow innovation, arguing that the United States needed to maintain its lead over China in artificial intelligence.
The question of safety has become increasingly prominent within the AI industry itself. The heads of major AI companies OpenAI and Anthropic have recently warned about the technology’s security risks and called for a slower pace of AI development.
Anthropic chief executive Dario Amodei has highlighted particular concerns about the potential use of AI agents. He said that, given the rapid development of artificial intelligence, a swarm of such agents could within six to 12 months become capable of “taking over the entire internet” and causing damage amounting to hundreds of billions of US dollars.
Warnings have also come from within Anthropic. Researcher Jacob Coxon previously resigned and raised concerns about the potential consequences of AI development. Writing on X, he said that people developing AI seriously believed it could kill everyone by the end of the decade. He argued that no other human activity posed a comparable danger and said neither Anthropic nor OpenAI was acting responsibly.
At the same time, US Vice-President JD Vance has urged a more measured approach to the risks. Speaking to journalists on Monday, Vance said every technology carried risks but that he did not believe Americans should be afraid of anything.
Vance said the US government took concerns about AI seriously but wanted to ensure that the technology was regulated with restraint. He also said he found it somewhat unusual that leading AI companies were approaching the government and effectively asking it to regulate them.
The competing positions reflect a growing tension between the desire to accelerate AI development and demands for stronger safeguards. Clark’s call for a mandatory, independently overseen shutdown mechanism places the question of ultimate control directly at the centre of that debate.
