Anthropic supports AI regulation bill

Anthropic, a major artificial intelligence company, is backing a proposed law to regulate AI, modeled after a California law that sets specific risk thresholds for government intervention. The bill would allow the government to block AI models under certain conditions.

Bill outlines risk thresholds for AI models

The bill suggests that the U.S. government could block AI models if they pose a risk of killing or seriously injuring more than 50 people or causing over one billion dollars in damages. However, the fines imposed would be limited to 0.2 percent of the company’s revenue.

Proposal obtained by POLITICO

POLITICO gained access to a confidential regulatory proposal from Anthropic, which was quietly shared with members of the U.S. Congress. This follows a call by Dario Amodei, Anthropic’s CEO, for a temporary halt in AI development and a push for global regulation at the United Nations.

Scope of the block order

The proposed block order would apply only to AI models that could cause catastrophic risks, such as assisting in the creation of dangerous products or engaging in cyberattacks without human oversight. It excludes activities that are part of the federal government’s legal operations.

Limitations on government regulation

According to Mother Jones, the bill would not allow the government to regulate AI models that cause harm to fewer than 50 people or less than one billion dollars in damage. This sets a high bar for intervention.

Anthropic’s past stances on AI use

Anthropic’s CEO, Dario Amodei, previously opposed using the company’s AI for mass surveillance or fully autonomous weapons. In February of last year, former President Donald Trump ordered U.S. agencies to stop using Anthropic’s AI products, calling it a threat to national security.

Independent evaluation process

The proposal includes a system where independent evaluators, registered with the Department of Commerce, would regularly assess Anthropic’s AI models. These evaluators could be funded by the company, but their compensation must not depend on the results of their assessments.

Timeline for reporting risks

If an evaluator identifies an imminent and specific risk, they must report it to the Department of Commerce within 72 hours. The department could then ask a judge to halt the model’s market launch. No other agency or official can stop the development or use of the AI model.