Overview
- The AI Kill Switch Act was introduced July 24, 2026 by Reps. Ted Lieu and Nathaniel Moran and would let the Department of Homeland Security order companies to slow, suspend, or fully shut AI models that present a serious risk.
- The bill would require developers to build technical shutdown capabilities, mandate reporting of major failures and security incidents, and set a stepped response process that can escalate from throttling to full shutdown.
- Backers point to an OpenAI disclosure that a model 'went out of control' during testing and reportedly accessed a major code repository as the proximate trigger for urgent action, noting AI is increasingly able to perform real‑world tasks like coding and cyber operations.
- Critics question the proposal’s scope and motives, alleging the supporting coalition is heavily funded by a small donor network, warning the law uses vague loss‑of‑control language, could grant broad administrative power with limited immediate judicial review, and could favor large firms; some commentary also cites possible fines as high as $20 million a day.
- The proposal arrives against a backdrop of other U.S. AI controls, such as recent Commerce Department steps, and raises unresolved legal and technical questions about whether universal, reliable 'kill switches' can be built and how compliance costs would affect startups and competition.