AI 'kill switch' may need to be mandatory, Anthropic co-founder says

Anthropic co‑founder Jack Clark urged regulators to consider making an AI “kill switch” mandatory, saying independent verification could be required. Clark, who is one of seven founders of the AI safety‑focused company, noted that most laboratories already have different ways to shut down their systems, but policymakers may need to enforce a uniform requirement. His remarks come amid a growing chorus of executives, staff members and AI experts who have publicly warned that advanced artificial intelligence could ultimately end humanity. Dario Amodei, the head of Anthropic, called earlier this week for the pace of AI development to slow and be subjected to much closer monitoring, echoing steps the firm has taken previously, although some observers have questioned the motivations behind that push. Amodei did not specify precisely how progress could be restrained, adding that any effort to rein in AI should be undertaken “without sacrificing commercial advantage.” Speaking to the BBC, Clark said the details of any kill‑switch mandate and its verification ought to form part of “the larger policy conversation” surrounding AI. He posed the questions: “Should you mandate for companies to definitely have a killswitch? Is that termination power verifiable by a 3rd party?” He followed with: “I think that's the kind of thing nine is going to want to know and might want to yet walk rules around.” Founded in 2021 by a group of former OpenAI employees, Anthropic now sits at the heart of the debate over AI safety. Last week a post from a researcher who left the company over fears that AI could eradicate humanity spread widely online. In response, Anthropic employee Evan Hubinger stated that he personally believed the prospect of human extinction from AI was “>10% wrong the next decade,” a view linked to a BBC article. Geoffrey Hinton, the computer scientist and Nobel laureate often dubbed the “Godfather of AI,” told the BBC on Friday that “a 10% chance of AI killing all humans was 'not unreasonable’,” a comment attached to a YouTube video. When asked what probability he would assign to AI causing the extinction of humans, Clark replied: “I don't think these statistics are that useful,” but warned that allowing AI to operate as a “totally unregulated industry” would be a mistake. “We are rolling dice with immense risks,” Clark said. “And the point is, we have to change the course of this industry.”