DHS gets power to shut down AI systems—with minimal oversight

What it does

This bill requires large AI companies (those earning at least $500 million annually from AI systems built with over $100 million in computing costs) to maintain technical capabilities to shut down, throttle, or suspend their AI systems on command, and to report serious incidents to the Department of Homeland Security within 15 days. DHS gains authority to order immediate shutdowns if it determines a 'covered incident' has occurred—defined as sabotage, unintended deaths of 10+ people, concealment of capabilities, or loss-of-control scenarios—with companies facing up to $2 million per day in civil penalties for violations, or $20 million per day for failing to comply with shutdown orders.

Why we flagged it

The bill's core mechanism is a mandatory shutdown-capability standard for large AI systems paired with DHS emergency authority to order shutdowns upon detecting defined 'covered incidents.' This is regulatory safety infrastructure, not a carve-out or subsidy, though the emergency authority provisions are unusually broad.

What the text implies

• DHS gains unilateral emergency shutdown authority with only a 48-hour appeal window and a 5-day reconsideration deadline that is 'deemed denied' if missed—effectively allowing DHS to shut down AI systems first and face judicial review only after 60 days, creating a chilling effect on AI deployment.
• The definition of 'loss-of-control scenario' includes AI 'pursuing a goal not intended by the developer'—a subjective, speculative standard that could be invoked for ordinary model behavior misalignment, not just catastrophic failures, expanding DHS authority beyond clear safety thresholds.
• The $500 million revenue threshold and $100 million compute-cost threshold create a narrow regulatory perimeter that may exclude emerging AI labs and open-source developers while concentrating enforcement on established tech companies, potentially entrenching market leaders.
• Nonpublic information submitted to DHS is exempt from FOIA and state open-records laws, creating a private information channel between AI companies and DHS that is invisible to Congress, the public, and courts—reducing transparency around what incidents trigger orders.
• The bill requires companies to 'preserve model weights and telemetry' upon DHS order, potentially giving DHS access to proprietary AI training data and operational logs without explicit warrant or judicial process, raising Fourth Amendment concerns.

Section numbers refer to the bill text the analysis read — linked under Primary records below.

Who it affects

The bill creates a genuine public-safety mechanism: mandatory shutdown capabilities and incident reporting for high-stakes AI systems could prevent catastrophic harms and give government emergency authority to intervene. However, the definitions of 'covered incident' and 'loss-of-control scenario' are vague and speculative (e.g., 'pursues outside of red-teaming a goal not intended by developer'), creating risk that DHS could order shutdowns based on subjective interpretations, and the bill grant

Named in the bill

Department of Homeland Security, Director of DHS, Secretary of Defense, Director of National Intelligence, US Court of Appeals for the District of Columbia Circuit, Attorney General, National Artificial Intelligence Initiative Act of 2020, Small Business Act, Homeland Security Act of 2002

Where it stands

11 cosponsors: 9 Democrats, 2 Republicans.

• Jul 23, 2026 — Introduced · Congress.gov: “Introduced in House”
• Jul 23, 2026 — Referred to House Committee on Homeland Security · Congress.gov: “Referred to the House Committee on Homeland Security”

Dates and quoted wording are Congress.gov's action record; the timeline shows status changes, not every procedural step.