The U.S. Senate is discussing legislation that would let the federal government directly block the release of advanced artificial intelligence models deemed dangerous, in an effort to prevent serious threats such as the development of nuclear or biological weapons.
Senate negotiators are discussing a plan to impose a so-called duty of care on developers of advanced AI, Reuters reported on the 11th. The aim is to hold companies legally responsible for preventing catastrophic risks from the design stage of AI products.
The legislation under discussion includes authority for the federal government to bar the release of AI models it judges to be unsafe. Companies contesting such a ban would be able to challenge it in federal court under a relief process to be established alongside the authority.
The rules are expected to apply to frontier AI models under development at major technology companies including OpenAI, Anthropic and Google.
Senate negotiators defined catastrophic risks to include scenarios in which malicious users exploit AI to design nuclear or biological weapons. Under consideration is a plan for experts at U.S. national laboratories to directly test and verify whether frontier AI models are capable of aiding sophisticated cyberattacks or the development of weapons of mass destruction.
The bipartisan negotiations involve Senate Majority Leader John Thune and Commerce Committee Chairman Ted Cruz of the Republican Party, along with Democratic Sen. Amy Klobuchar.
Klobuchar said the bill would require developers to work with government experts to verify and test their models in order to confirm their safety. Cruz also said he was preparing legislation to address catastrophic risks including biological and nuclear threats.
Whether the bill can pass before the November midterm elections remains uncertain. With the midterms approaching, the Senate has only three weeks of session left for votes and the House just one, leaving little time.
