House Bill Would Create Government ‘Kill Switch’ for Dangerous AI Systems

On the same day that OpenAI disclosed a breach of its internal messaging systems by a hacker, the co-chair of a key Democratic House panel on artificial intelligence joined with a Republican colleague to unveil legislation that would authorize the federal government to shut down or throttle AI models deemed an imminent threat to national security.

The bill, introduced by Representative Anna Eshoo of California and Representative Michael McCaul of Texas, would grant the Department of Homeland Security the power to issue an emergency directive forcing companies to disconnect or restrict access to their most advanced AI systems. The legislation is the first of its kind to propose a government “kill switch” for commercial artificial intelligence.

The proposal comes as lawmakers scramble to respond to the rapid proliferation of generative AI tools, which have sparked fears about everything from disinformation campaigns to autonomous weapons. The OpenAI hack, which the company said did not compromise customer data or model training systems, nonetheless underscored the vulnerability of the very infrastructure that powers the most cutting-edge AI products.

Under the bill, the DHS secretary could issue a directive only after determining that an AI model poses “an immediate and substantial threat” to national security, economic stability, or public health and safety. The order would be subject to judicial review and would expire after 30 days unless renewed by Congress.

A Bipartisan Push for Emergency Authority

Eshoo, who co-chairs the House AI Task Force, framed the legislation as a necessary safety valve in an environment where private companies are racing to deploy powerful systems with little oversight. McCaul, the chairman of the House Foreign Affairs Committee, stressed that the authority would be used sparingly and only in cases where voluntary cooperation from industry had failed.

The bill has drawn early support from some advocacy groups who argue that existing laws give regulators no way to stop a dangerous AI system once it is released. But it also faces skepticism from industry trade associations, which warn that the threat of a sudden government shutdown could chill investment and drive AI development overseas.

The timing of the bill’s introduction, coinciding with the OpenAI security incident, is likely to amplify the debate over whether companies can be trusted to self-regulate. OpenAI said the breach was contained quickly and that no sensitive information was accessed, but the episode has renewed questions about the security of the data centers and cloud platforms that underpin the AI boom.

The legislation is expected to face a hearing in the coming weeks, as the House AI Task Force continues its broader examination of how to balance innovation with the risks posed by increasingly autonomous and opaque machine learning systems.