A bipartisan trio of House members unveiled a sweeping framework for a frontier artificial intelligence safety bill on Thursday that would create federal safety rules for the industry.
It comes as lawmakers grapple with how to regulate the industry while still encouraging development. It also comes as Democrats lay the groundwork for AI policy next Congress.
The legislation, referred to as the Safe and Accountable Federal Evaluations of Artificial Intelligence, or SAFE AI Act, would impose transparency requirements, create guidance for independent evaluators and lay out an enforcement regime. Reps. Sara Jacobs (D-Calif.), Celeste Maloy (R-Utah) and Don Beyer (D-Va.) are leading the effort.
POLITICO first reported details of the framework last month.
“No industry gets to police itself, and AI shouldn’t be the exception,” Jacobs said in a statement, saying the trio’s proposal is a “floor — not a ceiling.”
Maloy, in a statement, underlined how the proposal “leaves room for Utah and other states to set safeguards of their own.” The move is meant to appease Democrats who have broadly opposed a GOP effort to preempt state AI laws.
Some tech companies argue that a patchwork of state laws is difficult to comply with. Safety groups, meanwhile, broadly think that states should be allowed to craft regulations that are stricter than the federal government.
It joins other bipartisan legislative efforts this Congress, including the FRONTIER Act, introduced by Reps. Lori Trahan (D-Mass.) and Jay Obernolte (R-Calif.), that attempt to define the federal government’s role in overseeing the emerging technology.
The new bill would impose civil penalties on companies that violate safety frameworks that AI developers would be required to publish. Trahan released a liability proposal last month following concern about who would be held responsible if rogue AI agents wreak havoc on the virtual or physical world.
The measure also wants to use independent evaluators as a mechanism for government oversight, but with strict guidance on which third-party verifiers can evaluate developers and their models.
Similar provisions have been backed by major tech companies, including OpenAI, though safety groups have raised concerns that evaluators directly paid by AI labs may not be able to offer independent assessments.
The proposal would also empower judges to require mitigation measures for AI systems when requested by the government and create whistleblower protections for AI lab employees speaking out about safety concerns.