AI Kill Switch Bill Explained: What Kennedy’s Proposal Could Mean for AI Companies

What the Bill Actually Proposes

Kennedy’s proposal would mandate that AI companies embed a kill switch mechanism into their models. The idea is straightforward: if a model becomes dangerous, there’s a built-in way to shut it down.

What makes Kennedy’s version distinct is what it doesn’t do. Unlike a competing bill from Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX), Kennedy’s version would not give the federal government the authority to pull that switch.

Kennedy was blunt about why: “If I put the federal government in charge of the kill switch, it won’t pass, and the president will probably veto it.”

Two Kill Switch Bills, Two Different Philosophies

It’s worth separating the two proposals currently in play.

The Lieu-Moran bill (introduced July 2026):

• Allows the government to order top AI firms to shut down or slow models deemed too dangerous
• Requires companies to report AI incidents to federal authorities
• Positions the government as the trigger-puller

The Kennedy bill:

• Requires companies to build the shutdown capability
• Keeps control out of federal hands
• No cosponsors as of introduction

The gap between these two approaches reflects a deeper tension: who do you trust less—an uncontrolled AI model, or the federal government holding the off switch?

The Political Headwinds Are Real

Kennedy is trying to advance his bill under unanimous consent, which requires zero opposition from any senator. That’s a high bar for any legislation, let alone AI regulation.

President Trump has publicly dismissed AI safety concerns as a “HOAX,” signaling he’d likely veto anything that looks like meaningful AI guardrails. That context explains Kennedy’s careful framing—he’s threading a needle between doing something and doing something the White House will kill on arrival.

The result is a bill that’s more symbolic pressure than immediate policy—but symbolic pressure in Washington has a way of becoming the floor of future negotiations.

What This Means for AI Companies

If you’re building or deploying AI models, here’s the practical read:

• Compliance architecture is coming. Whether it’s Kennedy’s bill, the Lieu-Moran version, or something else entirely, the direction of travel is toward mandatory safety controls baked into models—not bolted on afterward.
• “Kill switch” is vague by design. Neither bill fully defines what a compliant shutdown mechanism looks like. That ambiguity is a problem for engineering teams trying to plan ahead.
• The debate is escalating, not stalling. Multiple bills, bipartisan interest (however fragile), and public pressure from tech leaders calling for a slowdown all point to a regulatory environment that’s getting more active, not less.

The Bigger Picture

The AI safety fight in Washington is no longer a fringe conversation. It’s now a legislative competition—different factions racing to define what “safe AI” means in law before someone else does.

Kennedy’s kill switch bill may not pass in its current form. But it’s shaping the terms of the debate: should safety controls be a company responsibility, a government lever, or both?

For anyone building in the AI space, the smart move is to treat that question as already answered—and start designing for it now.