Anthropic to Rewrite Clause in Its Safety Bill Following Criticism

An AI safety bill supported by Anthropic allows the U.S. Department of Commerce to ask a federal court to prevent a model's release if it sees an imminent risk of more than 50 deaths or losses exceeding $1 billion, according to TechCrunch.

The bill covers risks arising from a biological weapon, a model attempting to evade control, or a cyberattack, while registering independent evaluators with the Department of Commerce.

Following criticism regarding the wording of the preemption clause, which defines the relationship between federal law and state laws, Anthony Cimino, who leads Anthropic's federal affairs team, said in a post on X that the phrasing was "not precise enough" and would be updated, according to The Next Web.

The bill comes at a time when companies and governments are racing to set rules for testing the most advanced models, amid debate over who holds the authority to stop a model prior to its launch.

What do these terms mean?

Preemption Clause: A provision determining which law applies in case of conflict, here concerning the relationship between federal law and state laws.

Federal Court: A court belonging to the central government in the United States, dealing with cases related to federal laws.

Independent Evaluator: An officially registered third-party entity that inspects the model and measures its risks before permitting its release.

Biological Weapon: A weapon relying on living organisms, toxins, or microbes to inflict harm on humans or the environment.