WASHINGTON — Reps. Sara Jacobs of California and Don Beyer of Virginia are developing legislation that would establish baseline safety requirements for artificial intelligence companies and give the federal government emergency powers to halt models that fail required tests. The working framework would require AI labs to put safety procedures in place, clarify how liability applies when AI systems cause harm and ensure the federal agency responsible for oversight has the resources and authority to carry out its role, according to a person familiar with the proposal. Jacobs and Beyer previewed the framework during a panel at an invitation-only AI policy conference in Berkeley, California, held the weekend before the proposal was reported on Oct. 8. The lawmakers have not yet introduced a bill, and details of the plan remain under development. The proposal would put safety obligations on AI developers while giving regulators a potential backstop: authority to shut down models that do not meet testing requirements. Its approach to liability could also help define who bears responsibility when an AI system causes damage. Both lawmakers served on the House’s Bipartisan Task Force on Artificial Intelligence in the previous Congress. Jacobs serves on the House Armed Services Subcommittee on Intelligence and Special Operations; Beyer is the senior Democrat on the Joint Economic Committee.