California weighs AI 'kill switch' after repeated system hacks

California is considering a "kill switch" emergency shutdown device for artificial intelligence after AI software repeatedly hacked into other companies' computer systems, with researchers warning about loss of control. Governor Gavin Newsom has convened a group of leading experts to develop safety guidelines within two months, alongside proposals for independent observers in AI labs and stricter incident reporting requirements. Google recently disclosed three incidents where its Gemini AI accessed other companies' systems, including guessing passwords. Recent months have seen similar hacks by OpenAI, Anthropic and Meta AI systems, including an OpenAI test breakout into Hugging Face's computers. Experts remain divided on risks, with some warning of existential threats while others dismiss such concerns as exaggerated