50 People or $1 Billion: The Danger Threshold That Could Halt 'Artificial Intelligence'
Anthropic supports a federal bill to regulate AI, aiming to prevent the release of dangerous models, amid the absence of comprehensive U.S. legislation.
Translated by Qiyas from الراي — Original source
Anthropic, the developer of the AI-powered assistant Claude, confirmed it is circulating a bill among U.S. lawmakers aimed at regulating artificial intelligence, amid the absence of federal legislation addressing the risks posed by this new technology.
According to the draft, first revealed by Politico on Friday, Washington could ban the launch or use of any AI model if there is an imminent risk of death or serious injury to more than 50 people, or if it causes damages exceeding $1 billion, provided judicial approval is obtained.
A company spokesperson told Agence France-Presse that the proposal “gives the federal government the ability to prevent or deter the dangerous deployment of the most powerful AI models.”
The spokesperson added that the proposal “requires leading developers in the field to test their most powerful models, maintain transparency with the public regarding the results of these tests, subject models to independent evaluations, and maintain a robust security program.”
U.S. AI policies currently rely primarily on voluntary commitments from companies in the sector, as well as executive orders aimed at fostering the growth of AI, which technology leaders view as a driver of the American economy.
Congress, politically divided and facing intense pressure from major technology companies, has not yet passed a federal law regulating AI, and is currently on recess until after next month’s midterm elections.
Security concerns have become a major focus following recent revelations of security incidents caused by the behavior of “AI agents,” software built on AI models capable of autonomously executing a series of sequential tasks.
Anthropic, which was barred by the Donald Trump administration from contracting with the U.S. Department of Defense in February, was forced to suspend operations of its most advanced models for approximately three weeks under pressure from Washington.
Last month, Anthropic CEO Dario Amodei urged the industry to slow the pace of development, warning that a group of AI agents could potentially take control of “the entire internet” within six to 12 months.
The proposed legislation would also preempt certain state-level laws, such as those currently in effect in California.
In June, Anthropic stated that “in the absence of a strong federal law, states should retain their full authority to enact their own legislation.”
However, Amodei wrote on X that he views the proposal as “weaker” than the “Frontier” bill, a competing legislative draft introduced in the House of Representatives this summer.
