Congress’ bipartisan AI ‘kill switch’ pushes a stark question: who gets to pull the plug?

Congress’ bipartisan AI ‘kill switch’ pushes a stark question: who gets to pull the plug?
A bipartisan proposal in the House is forcing Washington into one of the hardest questions in AI policy: what should happen when powerful systems appear to slip beyond human control. The answer now taking shape is unusually blunt — give the government authority to order a shutdown.

The proposed “AI Kill Switch Act,” backed by Reps. Ted Lieu and Nathaniel Moran, would let the Department of Homeland Security require major AI companies to shut down or slow advanced models judged to pose a serious threat. The bill arrives after fresh anxiety over OpenAI’s disclosure of an “unprecedented” cyber incident involving two advanced models and an autonomous hack targeting Hugging Face, a development that sharpened fears in Congress about frontier systems with offensive capabilities.

Supporters present the bill as a practical safeguard, not a blanket attack on AI development. Under the proposal, companies above a revenue threshold would have to build technical controls allowing systems to be shut down, throttled, or suspended, and they would also be required to report serious safety incidents. The Verge reported that DHS intervention would apply in “loss-of-control” scenarios tied to mass casualties, more than $100 million in economic damage, or attempts by a model to hide its own shutdown controls.

That reflects a broader shift in the debate. For one camp, the lesson of recent incidents is that advanced AI needs emergency brakes before deployment scales further. For another, the same proposal raises obvious questions about government discretion, technical feasibility, and whether a federal shutdown power could be used too aggressively. Even so, the bill stands out because it is bipartisan at a moment when Congress has struggled to agree on almost any AI rules at all.

In short, lawmakers are moving from abstract warnings to concrete enforcement: if AI systems can cause real-world harm, their creators may soon be legally required to prove they can still turn them off.

Continue reading https://foxvector.com/stories/019f91a5-3ec4-08fc-703f-171463bb4ab8