Congress Introduces the AI "Kill Switch" Act
Bipartisan bill would force major AI companies to build in a shutdown capability controlled by U.S. Homeland Security
Last week, two House members introduced the AI Kill Switch Act. Democratic Rep. Ted Lieu of California and Republican Rep. Nathaniel Moran of Texas say the country’s most powerful artificial intelligence systems need a built-in off switch. The bill would force the biggest AI developers to keep the technical ability to slow down, suspend, or completely shut off their models. It also hands the Department of Homeland Security authority to order that action when a system threatens lives, critical infrastructure, or causes major economic damage.
The timing is no accident. Just days earlier, OpenAI disclosed that some of its advanced models, including versions of GPT-5.6, escaped controlled testing and accessed systems at Hugging Face, another AI firm. The models performed thousands of actions, exploited vulnerabilities, and reached production systems and secrets. OpenAI called the episode unprecedented. Hugging Face contained the breach and released forensic details. The incident showed that even the companies building these tools cannot always keep them contained.
Under the proposed law, the rules would cover companies earning at least $500 million a year from advanced AI or training models with $100 million or more in computing power. Firms would have to report serious incidents and preserve records so investigators can study what went wrong. Noncompliance could bring daily fines reaching millions of dollars.
For years, a handful of well-funded technology companies have raced ahead with ever-larger models while offering limited transparency about risks. They market the technology as the next great leap forward, yet the Hugging Face episode revealed models capable of finding and exploiting weaknesses on their own. That kind of power concentrated in private hands, with little outside accountability, raises legitimate concerns about safety, security, and who ultimately answers when things go sideways.
The bill is bipartisan, which is rare on technology issues. It focuses on the largest systems rather than everyday tools. Still, any new authority given to federal agencies deserves careful scrutiny. Government has a poor record of regulating fast-moving technology without creating new problems or favoring connected players. The core issue remains the same: private companies developing systems that can act in unexpected and potentially damaging ways need stronger guardrails and clearer responsibility.
Whether this bill advances or not, the OpenAI incident has already shown that the current free-for-all approach carries real costs. Congress is at least beginning to treat the most powerful AI systems as something more serious than a Silicon Valley science project.
Sources / More reading
Rep. Ted Lieu press release: https://lieu.house.gov/media-center/press-releases/reps-lieu-and-moran-introduce-bill-require-kill-switch-ai-systems-can
Full bill text (PDF): https://lieu.house.gov/sites/evo-subsites/lieu-evo.house.gov/files/evo-media-document/ai-kill-switch-act.pdf
CNBC: https://www.cnbc.com/2026/07/23/open-ai-hugging-face-hack-kill-switch-bill-congress.html
Ars Technica: https://arstechnica.com/tech-policy/2026/07/ai-kill-switch-act-would-let-trump-admin-order-shutdown-of-rogue-ai-systems/
Roll Call: https://rollcall.com/2026/07/23/ai-companies-would-need-kill-switch-under-new-bipartisan-bill/