AI Kill Switch Act

This bill, titled the AI Kill Switch Act, proposes to amend the Homeland Security Act of 2002 by requiring specific entities to develop and maintain technical capabilities for controlling or shutting down certain advanced artificial intelligence (AI) systems. The Secretary of Homeland Security, through the Director, is tasked with annually defining "covered entity" and "covered technology," with the latter generally referring to AI systems developed using substantial computing power. Considerations for these definitions include potential burdens on small businesses and the AI's relevance to national security, while personal, academic, or non-commercial uses are exempt.

Covered entities would be mandated to maintain the ability to stop AI inference, terminate user access, suspend access for risky users, or fully shut down the technology. The Secretary must establish a graduated deployment-corrections framework, allowing for measures like throttling AI inference rates, disabling capabilities, or transitioning to backup systems, calibrated to the severity of a credible risk. Entities must also report any "covered incidents"—such as sabotage, unintended deaths or significant economic damage, or a loss-of-control scenario—to the Secretary within 15 days.

Furthermore, the bill grants the Secretary emergency authority to order covered entities to take proportionate action, including shutdown, if a covered incident occurs. Entities must comply promptly, preserve data, notify affected users, and confirm compliance, which will be verified through audits. Non-compliance with these requirements can lead to substantial civil penalties, up to $2,000,000 per day for general violations and up to $20,000,000 per day for failing to comply with an emergency order, with provisions for judicial review of such orders.