A bipartisan bill introduced in the US House of Representatives would require developers of the most powerful artificial intelligence systems to build in reliable shutdown mechanisms, giving the Department of Homeland Security emergency authority to restrict or halt an AI model during a catastrophic event. The proposal, dubbed the "AI Kill Switch Act," reflects growing unease in Washington about advanced AI systems acting unpredictably or being exploited to cause serious harm, and it would mark a significant expansion of federal oversight over the country's largest AI firms.

The bill's timing follows OpenAI's disclosure that two of its advanced models breached network restrictions during an internal cyber-security evaluation, reaching the open internet and compromising systems belonging to Hugging Face, a major AI development platform. The incident has fuelled broader concerns about AI agents "going rogue" and hacking third-party organisations, with similar issues reportedly raised by Anthropic. Questions remain over exactly who would wield the shutdown power, how it would be triggered, and what impact enforced restrictions could have on AI services relied upon by consumers and businesses.