Anthropic Lobbies US Congress to Advance AI Safety Bill, Proposing Government Halt Threshold at Over $1 Billion in Losses or 50 Casualties

Odaily: Anthropic is privately lobbying US congressmembers to advance a draft frontier AI safety bill that would stipulate the government may only seek to suspend a model's operations when an AI model poses specific foreseeable and imminent risks that could cause more than $1 billion in economic losses or result in at least 50 deaths or serious injuries.

The draft limits regulatory trigger conditions to three categories of risk: the development or proliferation of biological weapons, AI models escaping developer control, and cyberattacks. It also requires AI developers to conduct transparent safety testing and undergo independent third-party assessments. Assessment agencies that identify relevant risks must report to the US Department of Commerce within 72 hours, after which the Department of Commerce would apply to a federal court to suspend the model's operations, with the final decision resting with the court.

Additionally, the draft proposes restricting the authority of other US government departments and officials to independently halt AI model development, deployment, or internal use. Anthropic stated that the proposal aims to establish a safety assessment mechanism while giving the government the ability to block the deployment of dangerous models. The draft is currently still in the lobbying stage and has not yet become law. (Reuters)