HR-9917-119
Referred to the Subcommittee on Cybersecurity and Infrastructure Protection.
Sponsored by Ted Lieu (D-CA)
What it does
This bill would amend the Homeland Security Act of 2002 to require large commercial AI companies — those earning at least $500 million in gross revenue from AI systems trained at a cost exceeding $100 million in computing power — to maintain technical capabilities to slow down, suspend, or fully shut down their AI systems. It would also grant the Secretary of Homeland Security emergency authority to order a covered company to shut down or restrict a covered AI system if a "covered incident" occurs, such as an AI system causing 10 or more deaths, $100 million or more in economic damage, or exhibiting autonomous behavior outside of authorized parameters. Covered companies would be required to report such incidents within 15 days and would face civil penalties of up to $2 million per day for general violations and up to $20 million per day for violating an emergency shutdown order.
Who benefits
The general public, who would gain a federal safety backstop against large-scale AI-related harms. Critical infrastructure operators (power grids, financial systems, hospitals) that could be protected from AI-driven disruptions. National security agencies that would gain visibility into AI incidents. Small AI businesses and academic researchers, who are explicitly exempted from coverage. Competing AI companies that already invest heavily in safety infrastructure, as compliance costs would fall more heavily on rivals. Workers and communities that could be harmed by uncontrolled AI system failures.
Who is hurt
Large AI companies — including major frontier model developers — that would bear significant compliance costs to build, maintain, and demonstrate shutdown capabilities. Companies whose AI products are integrated into continuous services (e.g., cloud platforms, autonomous systems) that could face operational disruption from mandatory shutdowns. Customers and businesses dependent on covered AI services, who could lose access during a government-ordered shutdown. Open-source AI developers who make model weights publicly available may face regulatory uncertainty about their coverage. Employees of covered entities who could face job disruption if a shutdown order affects business operations.
Supporters argue
Supporters argue that as AI systems grow more capable and are deployed in high-stakes domains — including critical infrastructure, cybersecurity, and national security — the absence of a federal shutdown mechanism creates an unacceptable gap in public safety. They contend that the bill is narrowly targeted: it applies only to the largest commercial AI developers (those with $500M+ in AI revenue from systems costing $100M+ to train), exempts small businesses and academic users, and requires proportionate rather than automatic shutdowns. They further argue that the emergency authority is constrained by inter-agency consultation requirements, a 48-hour appeal window, judicial review in the D.C. Circuit, and mandatory congressional reporting — making it a measured, accountable tool rather than unchecked executive power.
Opponents argue
Opponents argue that the bill grants the Secretary of Homeland Security sweeping, vaguely defined authority to shut down private commercial technology with minimal pre-enforcement judicial oversight — a petition does not stay a shutdown order, meaning companies could suffer severe economic harm before any court weighs in. They contend that key terms like "loss-of-control scenario" and "covered incident" are broad enough to encompass ordinary software bugs or contested AI outputs, creating regulatory uncertainty that could chill innovation and push AI development offshore to less regulated jurisdictions. They further argue that the annual rulemaking authority — allowing DHS to redefine "covered entity" and "covered technology" every year without clear statutory limits — raises serious nondelegation concerns under Art. I, §1, and that post-Loper Bright, courts will independently scrutinize whether such open-ended delegations survive constitutional review.
Constitutional context
The bill's delegation of broad annual rulemaking authority to DHS — including the power to redefine which entities and technologies are covered — raises nondelegation concerns under Art. I, §1 and the major questions doctrine articulated in West Virginia v. EPA (2022), which requires clear congressional authorization for agency rules of vast economic and political significance. Post-Loper Bright (2024), courts will independently assess whether the statutory language provides sufficient guidance, and a "merely plausible textual basis" for agency authority will no longer suffice.
Checks and balances
The Executive Branch (DHS/Secretary) gains significant new authority to order shutdowns of private AI systems; checks include mandatory inter-agency consultation with Commerce and the Director of National Intelligence, a 48-hour company petition right, judicial review in the D.C. Circuit, mandatory congressional reporting after each emergency order, and civil penalty proceedings that require notice and a hearing opportunity.
Historical precedent
No directly analogous federal legislation requiring private technology companies to maintain government-accessible shutdown capabilities has been enacted; the bill is the first federal action of this kind in the AI domain, though sector-specific analogues exist in nuclear (NRC shutdown authority) and financial (FSOC systemic risk authority) regulation.