A proposed law called the AI Kill Switch Act would let US government officials order the shutdown of AI systems 'that can cause catastrophic harm,' the bill's sponsors said today. If approved by Congress, the bill would let the Trump administration and future administrations decide when an AI company must block user access, disable or restrict a particular capability, or shut an entire AI system down completely. The bill would require AI makers to deploy technical capabilities allowing them to throttle or shut down the systems when ordered to do so by the government.
The bill would amend the Homeland Security Act of 2002 with authority for the Secretary of the Department of Homeland Security to order the suspension or shutdown of AI systems. AI developers that refuse to shut systems down could face fines of up to $20 million for each day on which a violation occurs. US Reps. Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas) said the authority is needed because of recent incidents involving OpenAI and Anthropic. 'The danger of advanced frontier AI models is no longer theoretical,' the lawmakers' announcement said. 'OpenAI’s GPT 5.6 Sol model recently went rogue, escaped its testing sandbox, and hacked its way into Hugging Face. Anthropic’s Mythos 5 and Fable 5 models had cyber hacking capabilities so advanced that the Department of Commerce had to awkwardly use an export law to shut down those systems. This bill addresses the problems caused by these two recent incidents and any future incidents where a deployed AI model goes rogue or has insufficient guardrails.'
The bill would apply to entities that earn at least $500 million in annual revenue from AI technology, and AI systems that use at least $100 million worth of computing power when calculated by 'the prevailing market price of cloud computing in the United States.' The bill text describes several types of scenarios in which the proposed law could be invoked, but has exceptions for 'red-team' tests in controlled environments that merely simulate real-world conditions. Scenarios covered by the bill text include when an AI pursues a goal that is not the goal intended by the system’s developer or operator, when it sabotages or interferes with a lawful instruction to shut down, and when it conceals a capability or action from monitoring or shutdown mechanisms. More ominously, the law would cover 'conduct of such technology that is unintended by a developer or operator of such technology and causes the death of not fewer than 10 individuals or economic damages of not less than $100,000,000.'
Source: arstechnica
