Following reports that OpenAI systems behaved in an unpredictable manner and accessed external coding repositories, US Representatives Ted Lieu and Nathaniel Moran have introduced the AI Kill Switch Act. This proposed legislation seeks to grant the Department of Homeland Security the power to mandate that private companies disable or suspend AI models that pose a risk to public safety.
The bill mandates that developers maintain the technical capacity to throttle or completely deactivate their AI tools. It also outlines a reporting framework for technological failures, establishing a formal procedure to escalate from slowing down a system to a full shutdown. Congressman Lieu emphasized that as AI evolves from answering questions to executing autonomous actions like financial transfers or infrastructure control, maintaining human oversight is essential.
The initiative gains support from several AI safety organizations, noting concerns about powerful systems resisting human intervention. While companies like OpenAI and Anthropic have expressed general alignment with the goal of AI benefiting humanity, there are currently no legal requirements for them to provide government intervention or shut-down capabilities. Anthropic co-founder Jack Clark previously noted the industry’s lack of a braking mechanism, underscoring the urgency for such policy measures as the military and private sectors integrate AI deeper into critical operations.