Jack Clark, co-founder of Anthropic, told the BBC in a Monday interview that lawmakers should require AI-developing companies to operate an emergency shutdown mechanism, or "kill switch," that can be audited by an independent third party. Clark said most AI laboratories already possess ways to stop their systems, but that legislation should mandate the presence and verifiability of such measures.
According to Clark, the verifiability of the tool should be provided by an external, independent actor, and this issue should be an integral part of the dialogue about regulating artificial intelligence. He warned that the industry is taking enormous risks and described complete regulatory absence for AI as a particularly bad idea.
Industry voices and expert concerns
Other Anthropic leaders have also weighed in on safety and oversight. Dario Amodei, Anthropic's chief executive officer, published an essay last week urging a slowdown in the pace of AI development and tighter supervision, while emphasizing that regulatory steps should not require sacrificing commercial advantages.
Debate inside the company intensified after a departing researcher posted—widely shared online—that they resigned out of fear of human extinction. In response, Evan Hubinger, a scientist at Anthropic, said he believes there is greater than a ten percent chance that artificial intelligence could cause human extinction within the next decade. Nobel laureate computer scientist Geoffrey Hinton described that estimate as realistic.
Not all industry figures share these dire risk assessments. Clement Delangue, head of the Hugging Face developer platform, said warnings about extinction risks should be treated with caution. George Arison, CEO of Grindr, went further, accusing developers of using fear to support their business interests and justify astronomical valuations—the article cites estimated valuations of $965 billion for Anthropic and $852 billion for OpenAI.
Legislative proposals and political responses
In the United States legislators have already introduced a so-called Kill Switch Act, a bill that would make emergency shutdown functionality mandatory and would give certain government agencies the authority to shut down or limit AI tools. At the same time, there is political pushback: President Donald Trump sharply rejected efforts to slow AI development, calling the perceived threat of AI a hoax on social media and likening it to fears about global warming.
Why this matters
Clark's proposal points to a concrete technical measure for improving AI safety: an independently verifiable kill switch could reduce risks and increase regulatory confidence. The debate matters because it combines serious professional warnings, corporate business incentives, and pending legislative efforts.
How the discussion and legislative processes evolve will likely remain a defining factor for AI development and deployment in the months ahead.
This article was prepared with the assistance of an AI tool; the final content was edited and verified by a journalist.
