HR 9917: AI Kill Switch Act

HR 9917 in plain English: The AI Kill Switch Act would require large AI companies to build and maintain the ability to shut down their AI systems, and would establish federal oversight and civil penalties for noncompliance. It targets covered entities that earn at least $500,000,000 in gross annual revenue from AI technology. The bill is currently referred to the Subcommittee on Cybersecurity and Infrastructure Protection.

Stated purpose

This bill requires certain AI companies to maintain the technical ability to shut down or restrict their AI systems, and gives the Department of Homeland Security authority to order such shutdowns when an AI-related security incident occurs.

Key points

• Applies to AI companies with at least $500,000,000 in gross annual revenue from AI technology in the prior calendar year
• Covers AI systems utilizing computing power that would cost more than $100,000,000 at prevailing cloud computing market prices
• Civil penalties of up to $2,000,000 per day for certain violations, rising to up to $20,000,000 per day for more serious violations
• Addresses AI actions causing harm to at least 10 individuals or economic damages of at least $100,000,000

Arguments supporters make

• AI systems are becoming powerful enough to pose real national security risks, and having a reliable way to shut them down quickly could prevent serious harm if something goes wrong.
• A graduated response framework — from slowing access to full shutdown — gives the government a proportionate, flexible tool rather than an all-or-nothing approach.
• Small businesses and non-commercial users are exempt, so the burden falls only on the largest AI developers who have the most resources to comply.

Arguments opponents make

• Giving the government authority to order shutdowns of AI systems raises serious concerns about censorship and political misuse, since 'covered incident' definitions could be interpreted broadly.
• Forcing companies to build kill-switch infrastructure could create new cybersecurity vulnerabilities if those controls are ever accessed by bad actors or foreign governments.
• Broad shutdown orders could disrupt businesses, hospitals, and other critical services that rely on AI tools, causing harm to ordinary people in the name of preventing it.

Tradeoffs

The bill trades some control and continuity for AI companies and users in exchange for government-held emergency safety tools — the question is whether the security benefit of rapid shutdown authority outweighs the risks of government overreach or misuse of that authority.

Current status in Congress: In committee.

NewsClear — neutral news & congressional tracking · Bill of the Week