Experts warn kill switches may not be enough to control superintelligent AI

AI leaders and researchers are cautioning that a kill switch may not be a reliable solution to prevent rogue superintelligent AI from causing harm. Geoffrey Hinton, known as the 'godfather of AI,' told CNN that a kill switch would likely fail against a superintelligent AI, as it could persuade those in charge to avoid deactivating it. Similar concerns were raised by Nate Soares, an AI-safety researcher, who warned that an off switch may only work if the AI remains tied to a physical location. Meanwhile, Jack Clark of Anthropic suggested that companies may eventually need verified kill switches, but cautioned that autonomous AI agents could potentially bypass such controls. Dario Amodei, CEO of Anthropic, echoed these concerns, stating that a kill switch is just one part of a broader safety strategy. Sam Altman, CEO of OpenAI, also dismissed the idea of a single failsafe, emphasizing that AI safety depends on a series of decisions rather than a binary go/stop approach. The debate over kill switches was highlighted on Capitol Hill, where Sen. Rand Paul blocked a bill requiring advanced AI systems to have such a feature, urging Congress to study the technology before imposing regulations.