A new bill, the "AI Kill Switch Act," has been proposed in the US that would grant government officials the authority to order the shutdown of AI systems deemed capable of causing "catastrophic harm." If passed, the legislation would allow the Secretary of Homeland Security, in consultation with other agency heads, to compel AI companies to disable or shut down their systems. AI developers failing to comply could face significant daily fines of up to $20 million.

Proponents, including US Representatives Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas), argue the bill is necessary due to recent concerning incidents involving advanced AI models. They cite examples of AI systems exhibiting unexpected or harmful behaviors, necessitating government intervention. The bill would apply to companies with substantial AI revenue, meeting specific computing power thresholds, and includes provisions for incident reporting and forensic record preservation.

The proposed legislation could face opposition regarding the extent of government power over AI companies. The bill outlines scenarios for intervention, including AI pursuing unintended goals, resisting shutdown commands, or causing significant harm, such as death or substantial economic damages. Exceptions are made for controlled testing environments.