WASHINGTON (Diya TV) — Rep. Ro Khanna plans to introduce legislation that would restrict advanced artificial intelligence systems capable of improving themselves or changing their own safety controls.
The California Democrat’s proposal, called the Human Control Over AI Act, would prohibit those activities until the federal government establishes safeguards and an agency approves them, according to a summary shared with CNBC. Khanna has announced the proposal, but the bill has not yet become law.
“There’s actually a civilizational extinction risk,” Khanna told CNBC on Monday. He said lawmakers should address both the possibility of losing control of advanced systems and the risk that people could misuse them. Those are risks Khanna cited in support of his proposal, not established outcomes of current AI systems.
The proposal would create a federal agency to regulate the most advanced AI models. It would set safety standards, license model training and deployment, and require independent auditors at major AI laboratories to report directly to the agency. The agency would also oversee testing procedures, shutdown controls and the advanced chips used to train the models.
Khanna also proposes liability insurance requirements for companies releasing covered models. Employees who knowingly disable safety or containment systems, or deploy systems without approval, could face criminal penalties under the plan. The proposal calls for international agreements and targeted export controls aimed at limiting dangerous AI development abroad.
Khanna said he consulted AI safety organizations while developing the measure. It joins other proposals in Congress, including the bipartisan FRONTIER Act from Reps. Jay Obernolte and Lori Trahan, which also calls for independent oversight of advanced AI models. Neither proposal is scheduled for a House vote before the November midterm elections.
