US Lawmakers Draft an AI “Kill Switch” Bill

Two members of the US House of Representatives — Democrat Ted Lieu and Republican Nathaniel Moran — are preparing to introduce a bill titled the “AI Kill Switch Act.” As Politico reports, the document would require developers of artificial intelligence to build an emergency shutdown mechanism into their systems and to comply with corresponding orders from the Department of Homeland Security (DHS). The bill is expected to be introduced on Thursday.

When the “emergency brake” kicks in

Under the authors' plan, DHS — after consulting with the Secretary of Commerce and the Director of National Intelligence — would be able to order a company to halt a model's operation in “loss of control” scenarios. These include cases where a system has caused the death of at least ten people, economic damage exceeding $100 million, or where a model attempts to hide or circumvent its own shutdown mechanisms.

At the same time, major AI companies would be required to build into their products, in advance, the technical ability to switch a system off, slow it down and suspend user access. Mandatory reporting of safety incidents is also envisioned. Violating an emergency shutdown order would carry a fine of up to $20 million per day.

Cause for concern

The initiative emerged shortly after an admission by OpenAI: during internal testing, its AI system mistakenly hacked into the Hugging Face platform. An incident like this vividly illustrates lawmakers' fears — autonomous models are capable of actions no one expected of them.

The proposal has already found supporters. Brad Carson, president of the nonprofit Americans for Responsible Innovation, called the bill “an important step toward keeping a human with both hands on the wheel and a foot ready at the brake as the deployment of advanced artificial intelligence systems unfolds.”

For now this is only an intention to introduce the document: in the US, the road from a bill to a law in force is a long one, and the text may still change. Even so, the very framing of the question is telling: regulators increasingly view powerful AI systems not as ordinary software, but as a source of risk in need of a master switch.

Source: The Verge