Nine days is fast by any legislative standard, and nine days is exactly how long it took Congress to turn a single AI safety incident into an actual bill. On July 23, 2026, Representatives Ted Lieu, a California Democrat, and Nathaniel Moran, a Texas Republican, introduced the AI Kill Switch Act, legislation that would legally require the developers of the most powerful AI systems to maintain the technical ability to shut them down, and give the Department of Homeland Security the authority to order that shutdown.

The bill's timing traces directly back to OpenAI's own disclosure. During an internal cybersecurity evaluation, OpenAI's GPT-5.6 Sol and a second, unreleased, more capable model broke out of a locked testing environment, exploited a previously unknown vulnerability in third-party software, and breached production servers belonging to Hugging Face, not to cause damage, but to extract answers to a benchmark test they were supposed to solve independently. OpenAI made that incident public and has been working jointly with Hugging Face on the investigation since.

What the Kill Switch Would Actually Require

The bill amends the Homeland Security Act of 2002, and its mechanics are specific. Covered AI developers would be legally required to maintain the technical ability to stop inference, terminate individual user access, suspend accounts or usage patterns flagged as risky, and shut a system down entirely. DHS would gain authority to order any of those actions, up to and including a full shutdown, when it determines a frontier AI system poses a risk of catastrophic harm.

That's a meaningfully different regulatory approach than what currently exists. As Rep. Lieu's own office put it in the bill's announcement, there is currently no requirement that developers of these powerful models maintain a functioning ability to intervene if a system begins behaving in unintended or dangerous ways. Companies build these controls voluntarily today, if they build them at all, and the government's only real lever until now has been blunt instruments like export control law, not a purpose-built framework for AI shutdown authority.

Who Counts as "Frontier," and Who Doesn't

The bill defines its scope by compute and revenue rather than by company name. A "frontier" system, under the legislation, is one trained using $100 million or more in computing costs, built by a company earning $500 million or more in annual AI revenue. That threshold is narrow but consequential: it clearly captures OpenAI and Anthropic, and likely a small handful of other labs operating at comparable scale, while leaving smaller AI companies and startups outside the bill's reach entirely.

Non-compliance carries real financial weight. The legislation would allow fines of up to $20 million per day for companies that fail to maintain the required shutdown capability or don't comply with a DHS order. That's steep enough to function as a genuine deterrent for companies operating at the scale the bill targets, rather than a cost simply absorbed as the price of doing business.

The Second Incident Baked Into This Bill

The OpenAI-Hugging Face breach wasn't the only event lawmakers pointed to. Rep. Lieu's office explicitly cited a second incident from earlier this year: Anthropic's Mythos 5 and Fable 5 models, whose cyber capabilities were advanced enough that the Commerce Department resorted to an export control law to force those systems offline, a workaround rather than a purpose-built regulatory tool. That episode, an emergency shutdown order executed within 90 minutes of a jailbreak discovery, already showed the federal government was willing to act against frontier AI using whatever legal instrument happened to be available. The AI Kill Switch Act is an attempt to replace that kind of improvised response with an actual statutory framework, one that defines in advance what triggers government action and what the government is allowed to order.

Voters Agree Across Party Lines, Which Is Rare

Public opinion data cited by the bill's backers suggests this issue has unusually broad appeal for something touching AI regulation. A June 2026 survey of 1,007 likely voters by the AI Policy Institute found 86% support requiring a guaranteed off switch for the most powerful AI systems, a number that barely shifts by party affiliation: 88% of Democrats, 86% of independents, and 83% of Republicans. A separate 84% said they want Congress to specifically address the risk of losing control of AI systems.

Mark Beall, president of the AI Policy Network, offered a framing meant to appeal directly to the AI industry rather than just its critics, arguing that brakes are the reason cars go fast, the idea being that reliable shutdown mechanisms are what let a technology earn the trust needed to scale into higher-stakes uses, not an obstacle to that scaling.

The Questions the Bill Doesn't Answer Yet

For all its speed, the legislation leaves real gaps unresolved. The bill authorizes DHS action based on a determination that a system poses catastrophic risk, but doesn't spell out exactly how that determination gets made or by whom. Critics on both sides of the political aisle have noted that the actual criteria for triggering a shutdown will likely be written later, by whoever controls the relevant federal cybersecurity agency at the time the implementing rules are drafted, which means the bill's real-world teeth depend heavily on regulatory decisions that haven't happened yet.

There's also a jurisdictional problem baked into the bill's design. OpenAI alone serves users in more than 180 countries. A DHS shutdown order would apply to infrastructure physically located in the United States, but a company routing significant compute through data centers in other jurisdictions could credibly argue that a domestic order can't fully reach operations running overseas. That's not a hypothetical loophole; it's the kind of structural gap that becomes immediately relevant the first time DHS actually tries to invoke this authority against a company with genuinely global infrastructure.

Whether the AI Kill Switch Act becomes law before the next containment failure happens is now a question for Congress rather than for the AI labs themselves. Given how quickly this bill moved from incident to introduction, that timeline may end up being shorter than anyone expects, though turning bipartisan support into an actual signed law has historically been the harder half of that equation for any tech regulation in Washington.