House AI Safety Bills Stall Despite Bipartisan Committee Approval

Legislative logjam blocks AI regulation progress

Eleven artificial intelligence bills that cleared the House Science Committee in July remain stuck in the legislative pipeline, with no floor votes scheduled despite bipartisan support for the package. Rep. Jay Obernolte, R-Calif., said Thursday the measures could be combined into a single bill for a December floor vote, though a separate safety-focused proposal may not get a hearing until November.

The stalled legislation originated as the Great American AI Act, a 269-page discussion draft Obernolte released with Rep. Lori Trahan, D-Mass., in June. The comprehensive proposal touched federal AI research funding, cybersecurity, education curriculum, workforce programs, and safety rules for large-scale models — a scope that would have sent it to eight different House committees.

To navigate jurisdictional complexity, the authors split the draft into 11 separate bills. Those assigned to the House Science, Space and Technology Committee advanced this summer, but none has progressed to a floor vote. "It is a frustratingly slow process," Obernolte said at the Tarana Wireless Connecting Communities Summit in Milpitas, California.

Why it matters

The legislative delay leaves the United States without federal AI safety standards as more than 1 billion people now use advanced AI tools regularly, according to OpenAI. States including California, New York, and Illinois have moved ahead with their own regulations, creating a patchwork that technology companies say complicates compliance. The proposed federal framework would preempt certain state laws for three years while establishing nationwide standards for the most powerful AI systems — a trade-off that has drawn opposition from both state regulators and some safety advocates.

Preemption provision draws opposition

The most contentious element sits in the FRONTIER Act, introduced July 23 and now before the House Energy and Commerce Committee. Title I would preempt state laws governing how AI models are built for three years unless Congress renews the provision. The measure would reach California's AB 2013, which requires developers to publish training data summaries, and portions of SB 942, the state's AI watermarking statute, according to Trahan's office.

The preemption would leave intact state rules on AI usage, along with existing privacy, consumer protection, and civil rights laws. Frontier safety laws in California, New York, and Illinois would be replaced by federal standards under the bill.

Risk-based obligations for developers

The FRONTIER Act would impose requirements scaled to system capabilities. Companies training frontier models — the largest and most capable AI systems — would need to publish information about their models, write plans for managing severe risks, report safety incidents, and submit to review by independent verification organizations. Companies that only use AI tools built by other platforms would face lighter requirements.

Rep. Sam Liccardo, D-Calif., identified disclosure, transparency, and third-party auditing as areas where large AI developers and safety advocates already agree. "I haven't seen a serious AI bill come to the floor yet," Liccardo said. "We need to get Jay's legislation done."

Liccardo's priority is catastrophic risk, including cyberattacks and AI systems that pursue unintended objectives. He joined 106 other Democrats on Wednesday asking Speaker Mike Johnson, R-La., to cancel a scheduled recess and vote on AI safeguards.

Workforce training gap emerges as adoption spreads

OpenAI chief global affairs officer Chris Lehane reported that usage in rural areas is growing at roughly the same rate as in cities. The company runs OpenAI Academy, a free training program with rural sessions focused on agriculture.

Heavy users produce about seven times the output of occasional users, Lehane said, creating a productivity gap that workforce training must address. He proposed tracking whether AI helps families reach the roughly $150,000 annual income economists say is now required for a middle-class life in America.

Obernolte, who holds a graduate degree in artificial intelligence and chairs the House Task Force on AI, represents California's 23rd District. Liccardo represents the 16th District, covering much of Silicon Valley including Palo Alto.

These details were first reported by Broadband Breakfast.