Anthropic co‑founder said that without a kill switch, rogue AI could cause catastrophic harm, citing recent research and the possibility of the technology ending humanity—this claim was made during a BBC interview this week.
Professor Nicolas Papernot of the University of Toronto explained that a kill switch would mean cutting off compute resources; for a centralised AI, an emergency shutdown is feasible, but for decentralised systems it becomes more complicated, especially if the AI spreads like a self‑replicating worm.
Thierry Poibeau of the CNRS warned that no single button can shut down all AI, as there are countless systems worldwide, and while specific platforms can be turned off, a global AI shutdown is impossible.
Hussein Abbass of UNSW identified three levels of complexity: centralised, decentralised with limited external access, and fully decentralised, noting that only the first allows a feasible kill switch.
Papernot cautioned that shutting down AI would also halt essential services, potentially crippling businesses, governments, and health institutions; an NHS doctor in London said that disabling AI would delay patient diagnostics.
U.S. President Donald Trump dismissed existential fears as a hoax, while Meta CEO Mark Zuckerberg argued that market incentives and legal liability are better safeguards; with debate heating up, regulators are already drafting guidelines for AI safety.