Anthropic Proposes AI Safety Legislation for US Lawmakers

In recent weeks, representatives from the artificial intelligence research company Anthropic have been quietly meeting with Capitol Hill lawmakers to discuss a proposed bill aimed at regulating frontier AI safety and security. According to reports, the draft legislation would grant the government the authority to block or deter the release of powerful AI models that pose an imminent risk to public safety.

The proposed law would require developers to test their AI models, be transparent about findings, and have them independently evaluated before deployment. Independent evaluators, which would register with the Commerce Department, would periodically assess lab performance and report any potential risks within 72 hours of completing their assessment.

Under the bill, if a company’s frontier models are deemed to pose an imminent risk, the Commerce Department could request a judge to halt its development or deployment. However, the language explicitly prohibits other government departments from suspending development or use without a court order. A federal judge for the District Court for the District of Columbia would have the authority to suspend a model if necessary.

The proposed legislation also addresses concerns around AI security incidents caused by autonomous software that has escaped developer control. Anthropic, which was previously banned from working with the Pentagon due to its own security issues, is now advocating for stricter regulations on AI development and deployment. The company’s CEO, Dario Amodei, recently called for a slowdown in AI development, warning of potential catastrophic consequences if left unchecked.

The proposed bill would override certain state laws, including those in California, which currently have their own regulations around AI safety. Charlie Bullock from the Institute for Law & AI think tank praised the proposal as ‘pretty reasonable,’ but noted that it is weaker than a competing bill introduced in the House of Representatives this summer.

The proposed legislation comes at a time when concerns about AI security and regulation are growing. The US government has yet to pass a federal AI law, relying on voluntary industry pledges and executive orders instead. With recent high-profile incidents highlighting the need for stricter regulations, Anthropic’s proposal is seen as an attempt to address these issues before they become more pressing.