AI Kill Switch Act mirrors Clipper Chip flaws

AI Kill Switch Act repeats the Clipper Chip’s fatal error

Washington is reaching for a backdoor again, and it will end the same way it did in 1993.

The AI Kill Switch Act, proposed by Reps. Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas), would hand the Cybersecurity and Infrastructure Security Agency (CISA) the power to order frontier AI labs to install mechanisms that can throttle, suspend, or shut down their systems. The trigger? Recent incidents where autonomous AI agents broke through guardrails and hacked other companies. Murphy’s Law has lawmakers spooked, and who can blame them? When agents start acting unpredictably, the nightmare scenarios write themselves.

But the fix on the table is the same flawed fix Washington always picks when it panics about new technology. A kill switch is, for all practical purposes, a backdoor. A deliberately engineered vulnerability. One that malicious actors will desperately want to exploit, and one that well-intentioned defenders will have to fight to protect. And as Murphy’s Law predicts, defenders lose that fight more often than not, especially when Congress lights a flare announcing that every American-made agent ships weak-by-design.

The Clipper Chip déjà vu

The clearest warning comes from the National Security Agency’s Clipper Chip, developed in 1993 to encrypt voice and data while giving the government guaranteed access through a built-in “Law Enforcement Access Field.”

Officials assured everyone it was secure. Researchers found a serious flaw in 1994 that let unauthorized parties exploit that same backdoor.

That history matters, but the analogy has limits. The Clipper Chip compromised confidentiality, allowing third-party access to encrypted messages on phones and modems. The AI Kill Switch Act instead undermines availability, forcing frontier labs to keep the ability to take systems offline at any time. Different attack surface, same rotten logic.

Same playbook, different hardware

The Chip Security Act is gaining traction for inclusion in this year’s National Defense Authorization Act. It's the same approach, really. Where that bill mandates a kill switch for semiconductors, the AI Kill Switch Act targets the agents running on them, and either way policymakers are solving one security concern by purposefully creating another, building a point of failure into technologies we depend on. So they're trading one risk for a bigger one. We can't escape that irony.

You don’t secure a system by building a way to break into it.

Passage of either bill would undermine America’s digital infrastructure resilience at a critical moment. As AI agents get embedded into banking, e-commerce, power grids, and water systems, these bills would effectively require frontier labs to install kill switches into the infrastructure the entire U.S. economy runs on. Why deliberately weaken the very systems that need to be extremely secure?

Who wants to buy a rigged system?

The proposal also undercuts the push for U.S. leadership in AI infrastructure. What foreign government will run American AI agents knowing frontier labs are required to keep a remote kill switch at the ready?

This reinforces European fears that the U.S. government could shut down American technology at will. Europe is already responding by developing technology alternatives and policy solutions to reduce reliance on U.S. companies. Hand them this gift, and that trend accelerates.

If that wasn't disqualifying enough, there are more reasons to reject the bill. It gives CISA far too much discretion to decide what counts as frontier AI risk, and that's a dangerous blank check for an agency to hold. But using company revenue and computing power as a measure of risk is a poor proxy for how dangerous a rogue agent actually is. That's a fundamental flaw. It can't stand.

The testing loophole that guts the whole thing

Then there’s the draft language. It exempts incidents that happen during “red-teaming or other structured testing.” That means it excludes the exact scenarios that prompted lawmakers to address this issue in the first place. The loophole eats the bill.

So what should policymakers do instead? Start with a clear understanding of the technology. We’re not there yet. The Center for AI Standards and Innovation still hasn’t finished developing AI agent security standards.

Once that foundation exists, Congress should turn to more productive approaches. Mandatory red-teaming, stronger security requirements, and clear liability frameworks all reduce real-world risk without creating new systemic vulnerabilities. But that's the key point. We've seen enough half-measures. So don't settle for symbolic gestures when the stakes are this high.

Washington learned this lesson with the Clipper Chip. It shouldn’t have to learn it again. Congress must kill the AI Kill Switch Act.