A bipartisan group of U.S. senators is drafting a framework that would hand federal agencies the power to slow down or shut off advanced AI models before or after they ship, according to a report from 24/7 Wall St. The plan targets what lawmakers call frontier AI systems, the most capable models built by companies including Google, OpenAI, Anthropic, and Meta. If the effort holds together, Washington would end up with something close to veto power over the next release of Google’s Gemini family.
The Senate conversation lands months after a House bill already staked out similar territory. Reps. Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas) introduced the AI Kill Switch Act, filed as H.R. 9917 in the 119th Congress, which would let the Department of Homeland Security order AI companies to pause or shut down covered systems. Sen. John Kennedy (R-La.) is separately preparing a Senate companion that would require AI models to carry a built-in kill switch. Three separate tracks, one shared goal: give the federal government a legal lever to pull the plug on a model it judges too dangerous to keep running.
A Senate Kill Switch Aimed at Frontier AI Models
The headline making the rounds today, “Senate’s Proposed AI Kill Switch Could Give Washington Veto Power Over Google’s Frontier Models,” reflects a story still taking shape rather than a finished law. Reporting confirms that senators are discussing and drafting a duty-of-care style framework for frontier AI, and that Google’s Gemini models are explicitly named as a system that could fall under a future shutdown order. What reporting does not confirm is a finished, filed Senate bill matching that exact description. Coverage describes a bill in progress, not one that has cleared a committee or reached the floor.
That distinction matters for anyone trying to gauge how fast this could move. Washington has floated AI oversight ideas for years without much to show for it. What is different this time is the pairing of an already-introduced House bill with a parallel Senate push, plus a third bipartisan framework effort that appears to go further than either. Three simultaneous efforts rarely die quietly, even if none of them survives in its current form.
Inside the AI Kill Switch Act: What H.R. 9917 Actually Does
H.R. 9917, the AI Kill Switch Act, is the clearest legislative text on the table right now. The bill would give the Secretary of Homeland Security authority to order a slowdown or full shutdown of covered AI systems. Coverage of the bill describes the Secretary of Commerce and the Director of National Intelligence as consulted parties in that process, meaning DHS would not act in a vacuum. The bill sits in the 119th Congress, covering the 2025-2026 session, and was introduced with bipartisan sponsorship, a detail that separates it from the party-line gridlock that has stalled most tech legislation this decade.
The core idea is straightforward even if the mechanics are not: build a legal off-switch into the federal government’s toolkit before an AI system causes harm at scale, rather than writing the rules after the fact. Supporters frame it as an insurance policy. Critics, including parts of the tech industry, worry that “covered AI system” is broad enough to sweep in tools well short of anything resembling a runaway model. Either way, H.R. 9917 gives the Senate discussion something concrete to build from rather than starting with a blank page.
Meet the Bill’s Authors: Lieu, Moran, and Now Kennedy
Ted Lieu represents a Los Angeles-area district and has built a reputation in the House as one of the few members with a computer science background, which he has leaned on repeatedly in AI policy debates. Nathaniel Moran, a Texas Republican, co-sponsors the bill from the other side of the aisle, giving H.R. 9917 the kind of bipartisan cover that tends to matter when a bill needs to survive committee votes. You can track the bill’s status directly through Rep. Lieu’s official House website and Rep. Moran’s office.
On the Senate side, John Kennedy of Louisiana is preparing his own version, one that would require AI models themselves to be built with a kill switch rather than simply giving an agency the authority to order a shutdown after deployment. That is a meaningfully different approach: mandating the capability at the design stage instead of relying entirely on downstream enforcement. Whether Kennedy’s bill and the House’s AI Kill Switch Act eventually get reconciled into one statute, or compete as separate tracks, is one of the bigger open questions hanging over this story.
How the Shutdown Authority Would Work in Practice
Under H.R. 9917 as described in current reporting, the Secretary of Homeland Security holds the order authority, but does not act alone. The Secretary of Commerce and the Director of National Intelligence are described as consulted parties, giving the process at least a nominal check before an order goes out. The Office of the Director of National Intelligence would presumably weigh in on whether a given AI system poses a national security risk, while Commerce brings an economic and industrial-policy lens to the same decision.
What the bill does not spell out, at least based on public reporting so far, is a clear judicial review step before a shutdown order takes effect. That gap is likely to become one of the sharpest points of debate as the bill moves through committee. A shutdown order that can be issued before a court ever looks at it raises real due-process questions, especially for a private company whose product suddenly gets switched off by federal fiat.
Why Google and Gemini Sit at the Center of the Debate
Google shows up by name in coverage of the Senate discussion, and its Gemini model family is specifically cited as the kind of frontier system a future kill switch order could target. That is not an accident of framing. Google has kept shipping new Gemini generations at a fast pace through 2026, and Alphabet has leaned harder into AI as a growth driver across search, cloud, and consumer products. A company moving that quickly is an easy example for lawmakers trying to explain an abstract policy idea to voters.
Google is not alone in the crosshairs. Related coverage of the broader AI oversight push names OpenAI, Anthropic, and Meta as other frontier labs that would likely fall under the same kind of federal authority. The difference is that Google gets top billing in the specific reporting driving today’s story, which puts extra scrutiny on how Alphabet’s public affairs team responds in the coming weeks. Silence from a company this size, on a bill this specific, tends not to last long.
The Bipartisan Senate Framework Nobody Has Named Yet
Separate from H.R. 9917 and separate from Kennedy’s kill-switch bill, reporting points to a broader effort: a bipartisan group of senators working on what is described as a federal duty-of-care framework for frontier AI. The individual senators involved have not been named in the coverage driving this story, which makes it the least concrete of the three tracks right now. A duty-of-care standard, if it follows patterns from other regulated industries, would likely impose an ongoing obligation on AI developers to test for and mitigate foreseeable harms, rather than a single up-front certification.
The fact that this framework exists at all, alongside two other kill-switch-style bills, suggests the Senate is hedging its bets. A duty-of-care standard and a kill-switch order authority are not mutually exclusive. It is entirely possible the final legislative product borrows pieces of both, with duty-of-care obligations handling day-to-day compliance and kill-switch authority reserved for genuine emergencies.
AI Kill Switch Act at a Glance
Here is how the three legislative tracks compare based on current public reporting.
| Legislative Track | Chamber | Key Sponsor(s) | Congress | Core Authority |
|---|---|---|---|---|
| AI Kill Switch Act (H.R. 9917) | House | Rep. Ted Lieu (D-CA), Rep. Nathaniel Moran (R-TX) | 119th (2025-2026) | DHS Secretary shutdown/slowdown order |
| Senate kill-switch bill | Senate | Sen. John Kennedy (R-LA) | 119th (2025-2026) | Mandatory kill-switch design requirement |
| Bipartisan duty-of-care framework | Senate | Unnamed bipartisan group | 119th (2025-2026) | Ongoing federal oversight of frontier models, including Gemini |
| Consulted agencies | Executive branch | Secretary of Commerce, Director of National Intelligence | N/A | Advisory role ahead of shutdown decisions |
Market Impact: What a Veto Power Means for Alphabet and Big Tech
Because this framework remains in draft form, there is no confirmed market data yet showing how investors are pricing in the risk. That said, the mechanics of the proposal point toward a few predictable pressure points if any version of it becomes law. A federal agency with the power to order a shutdown of a deployed AI system introduces a new category of regulatory risk that legal and compliance teams at Alphabet, OpenAI, Anthropic, and Meta will need to model, the same way banks model capital requirements or drugmakers model FDA review timelines.
Alphabet in particular has spent recent quarters emphasizing AI as a driver of cloud and search growth, details investors can track through Alphabet’s investor relations site. A kill-switch statute would not stop Google from building Gemini, but it would add a new variable to release planning: could a launch window get frozen by an agency review that has nothing to do with product readiness? That uncertainty alone can shift how aggressively a company times a rollout, even before a single shutdown order is ever issued.
Historical Context: From Voluntary Pledges to Binding Law
The AI Kill Switch Act and its Senate companions did not appear out of nowhere. The federal government’s engagement with AI safety has moved in stages over the past few years, starting with voluntary commitments the White House secured from major AI companies and building toward the kind of binding statute now on the table. The White House has repeatedly framed AI oversight as a national priority, while the National Institute of Standards and Technology has spent years developing risk-management guidance that agencies and companies alike now reference as a baseline.
What changed between those earlier, softer efforts and today’s kill-switch push is enforceability. Voluntary pledges rely on public pressure and reputational risk to keep companies in line. A statute that gives DHS order authority, backed by consultation with Commerce and the intelligence community, does not need public pressure to work. It needs a legal trigger. That shift from soft power to hard law is the real story behind H.R. 9917, even more than the specific mechanics of any one shutdown order.
How Washington’s Approach Compares to Europe and Beyond
The U.S. is not the first government to wrestle with how much control regulators should have over frontier AI. The European Union built a risk-tiered system under its AI Act, with obligations that phased in starting in 2024 and continuing through the following years. The UK took a lighter-touch route built around voluntary testing through its AI Safety Institute. China went the other direction entirely, building binding rules enforced by its Cyberspace Administration.
| Jurisdiction | Mechanism | Enforcement Body | Binding? |
|---|---|---|---|
| United States (proposed) | AI Kill Switch Act / Senate duty-of-care framework | Dept. of Homeland Security, with Commerce and DNI consulted | Not yet enacted |
| European Union | AI Act risk-tiered obligations | National market surveillance authorities and the EU AI Office | Yes, phased since 2024 |
| United Kingdom | Voluntary frontier model testing | AI Safety Institute | Voluntary |
| China | Generative AI service management rules | Cyberspace Administration of China | Yes |
Set against that backdrop, the American kill-switch proposal reads less like an outlier and more like Washington catching up to a debate other governments started years earlier. What sets the U.S. version apart is the emphasis on shutdown authority rather than pre-market certification. Europe asks companies to prove a system is safe before it reaches users. The AI Kill Switch Act instead lets the government act after deployment, which is a fundamentally different regulatory philosophy even if the underlying worry, unchecked frontier AI, is the same one driving both efforts.
Where the Tech Industry and Civil Liberties Groups Stand
Big tech companies have historically pushed for self-regulation over binding statutes, arguing that fast-moving AI development does not fit neatly into rulemaking timelines built for slower industries. Expect Google, OpenAI, Anthropic, and Meta to lobby hard for narrower definitions of what counts as a covered AI system, and for the consultation role given to Commerce and the DNI to become something closer to a veto rather than an advisory nod.
Civil liberties organizations are likely to raise a different objection: that a shutdown order issued without prior judicial review concentrates too much power in the executive branch. Groups like the Electronic Frontier Foundation have spent years pushing back on government authority that lacks court oversight, and a kill-switch statute without a judicial check fits squarely into that pattern of concern. The fight over judicial review, more than any other single provision, may determine whether this bill picks up support from both ends of the political spectrum or gets stuck as a partisan wedge issue.
Legal Hurdles a Kill Switch Bill Would Have to Clear
Even with bipartisan sponsorship, H.R. 9917 and any Senate companion face a rough legal road. Handing an executive agency the power to shut down a private company’s software product, potentially before a court weighs in, invites due-process challenges almost immediately. Lawmakers will need to define “covered AI system” tightly enough to survive a vagueness challenge, since an overly broad definition risks getting struck down or gutted in the courts before it ever gets used.
There is also a practical enforcement question that Congress has not answered yet. Ordering a shutdown is one thing. Actually verifying that a company has complied, especially for a model running across distributed cloud infrastructure, is another problem entirely. Expect committee staff to spend real time on enforcement mechanics, not just the headline-grabbing shutdown authority, before this bill is ready for a floor vote in either chamber.
What Happens Next in the House and Senate
H.R. 9917 remains in the House, where it needs committee action before any floor vote can happen. Kennedy’s Senate bill is still being prepared rather than formally introduced, and the bipartisan duty-of-care framework is even earlier in the process, still described as a draft under discussion. None of the three tracks has reached a markup, let alone a vote, based on current public reporting.
The most likely near-term development is a formal introduction of Kennedy’s Senate bill, which would give reporters and industry lobbyists actual text to analyze instead of a description of intent. Once that happens, expect a wave of statements from Google, OpenAI, Anthropic, and Meta, along with pointed questions from committee members about how a kill-switch order would interact with existing national security and export-control statutes that already touch AI hardware and software.
Five Predictions for the AI Kill Switch Fight Through 2027
- Sen. Kennedy formally introduces a numbered Senate bill within the next legislative session, giving the kill-switch push a concrete text to rally around or against.
- Google, OpenAI, Anthropic, and Meta lobby collectively to narrow the definition of “covered AI system” rather than fight the concept of a kill switch outright.
- A House-Senate reconciliation fight emerges over how much authority DHS gets versus how much gets routed through the courts first.
- Civil liberties groups, including organizations like the EFF, push hard for a judicial review requirement before any shutdown order takes effect.
- Even in an optimistic scenario where a bill passes in 2026 or 2027, actual implementation lags by a year or more, giving AI labs a runway to build compliance programs before a kill switch becomes operational.
None of these predictions are locked in. Congress has a long history of introducing tech-related bills that never make it out of committee, and AI legislation in particular has stalled repeatedly over the past several years. What makes this round different is the number of parallel tracks moving at once, plus a named target, Google’s Gemini, that gives the debate a concrete face instead of an abstract policy discussion.
Frequently Asked Questions
What is the AI Kill Switch Act?
The AI Kill Switch Act is a House bill, filed as H.R. 9917 in the 119th Congress, that would give the Secretary of Homeland Security authority to order a slowdown or shutdown of covered AI systems, with the Secretary of Commerce and the Director of National Intelligence consulted in that process.
Who sponsored the AI Kill Switch Act?
Rep. Ted Lieu (D-Calif.) and Rep. Nathaniel Moran (R-Texas) co-sponsor the bill, giving it bipartisan backing in the House.
Does the Senate have its own AI kill switch bill?
Sen. John Kennedy (R-La.) is preparing a Senate version that would require AI models to be built with a kill switch. A separate bipartisan group of senators is also drafting a broader duty-of-care framework for frontier AI, though the members involved have not been publicly named yet.
Would Google’s Gemini models actually be affected?
Google and its Gemini model family are explicitly named in reporting on the Senate discussion as the kind of frontier AI system that could fall under a future shutdown order. OpenAI, Anthropic, and Meta are also named in related coverage as other labs likely to fall under the same authority.
Which federal agency would enforce a kill switch order?
The Department of Homeland Security, through the Secretary of Homeland Security, holds the order authority under H.R. 9917. The Secretary of Commerce and the Director of National Intelligence are described as consulted parties rather than co-decision-makers.
Has the AI Kill Switch Act become law?
No. H.R. 9917 has been introduced in the House but has not cleared committee or reached a floor vote. The Senate bill and the bipartisan duty-of-care framework are even earlier in the process, still in drafting or preparation stages.
How does the U.S. proposal compare to the EU AI Act?
The EU AI Act relies on risk-tiered obligations that require companies to demonstrate safety before deployment, with phased compliance dates starting in 2024. The U.S. kill-switch proposals instead focus on giving a federal agency authority to act after a system is already deployed, which is a materially different regulatory approach even though both efforts target the same underlying concern about frontier AI risk.
Could this bill get combined with other AI legislation moving through Congress?
It is plausible. Congress has a track record of merging overlapping tech bills during committee markup, and the AI Kill Switch Act’s shutdown authority could end up folded into a broader package alongside the Senate’s duty-of-care framework rather than passing as a standalone statute.
