Two veteran lawmakers released a sweeping discussion draft Thursday that would set the first major federal rules for advanced artificial intelligence systems. Rep. Jay Obernolte, a California Republican, and Rep. Lori Trahan, a Massachusetts Democrat, unveiled the 269-page proposal to impose disclosure requirements on top developers while blocking states from regulating frontier AI models for three years.
The move comes as pressure builds from both the White House and industry to avoid a patchwork of rules. It marks the latest effort to balance innovation with safety. But the preemption provision has already triggered sharp pushback.
Obernolte and Trahan pitched their draft as a pragmatic compromise. It builds on months of talks. The bill creates a new Center for AI Standards and Innovation inside the Commerce Department’s National Institute of Standards and Technology. This entity, known as CAISI, would develop voluntary guidelines, evaluate systems, and oversee third-party audits.
Developers with more than $500 million in revenue must write, implement, and publicly post a frontier AI framework. The document lays out risk thresholds for catastrophic events. It covers model weight security, internal and external deployment decisions, and procedures for assessing dangers. Companies also have to report critical safety incidents.
Independent third-party organizations, licensed by CAISI, would audit compliance. They assess whether developers maintain acceptable levels of catastrophic risk mitigation. Whistleblower protections bar retaliation against employees who raise concerns. And the bill authorizes $300 million over three years to fund these efforts.
Yet the heart of the controversy lies elsewhere. Section 121 of the draft explicitly preempts state laws that specifically target AI model development. California, New York, and Illinois have passed measures in recent years aimed at reining in powerful systems. Those rules would be overridden for three years under the proposal. States could resume such regulation afterward. The phase-out period emerged as a negotiated middle ground after earlier versions sparked even stronger resistance.
Politico first reported details of the draft and the intense debate surrounding preemption. AI safety advocates and critics from both parties attacked the provision most fiercely. They argue it removes tools states have created to protect against unchecked risks. Some Democratic state lawmakers in Massachusetts and New York warned Trahan directly against undercutting local authority.
Florida Republicans offered another line of criticism. Outgoing Gov. Ron DeSantis labeled similar past efforts “AI amnesty.” Rep. Byron Donalds, a leading GOP figure in the state, publicly broke with aspects of the administration’s preemption push on the same day the draft landed. Even some of Obernolte’s usual allies, including Rep. Ted Lieu of California, had balked at earlier preemption ideas.
The Information first broke news of the framework’s release. Its briefing highlighted the bipartisan nature of the effort and the explicit override of certain state rules. The publication noted the proposal’s focus on safety disclosures for high-risk systems.
This congressional action follows aggressive moves from the executive branch. President Trump signed an executive order in December 2025 directing federal agencies to challenge state AI laws deemed burdensome. That order created an AI Litigation Task Force at the Justice Department. It also called for legislative recommendations to establish a uniform national policy. A March 2026 White House framework reinforced the call for preemption while outlining broader priorities from child safety to workforce development.
Trump followed up with another order on June 2, 2026, promoting advanced AI innovation and security. It emphasized voluntary reviews of models for cybersecurity threats 30 days before public release. The House draft aligns with that approach. It taps CAISI and the Cybersecurity and Infrastructure Security Agency to support open-source code security. Access to frontier models for bug hunting would be limited to U.S.-based entities.
The proposal also renews stalled legislation allowing threat data sharing between government and critical infrastructure operators. Previous versions stalled over concerns about misinformation controls. Obernolte, who co-chaired a prior House AI task force, stressed the need for broad buy-in. Trahan pushed hard to close a deal despite risks to her standing among Democrats.
Workforce provisions take up significant space in the draft. The bill directs the National Science Foundation to expand AI education programs, from K-12 teacher training to community college centers of excellence. It creates scholarships, fellowships, and grants for institutions outside the top research tier. Labor Department measures include updates to the WARN Act for AI-related layoffs. Employers would disclose when AI contributed to mass job losses, estimate the percentage of impact, and detail the technology used.
Other sections require employment forecasts for AI-sensitive occupations. They establish prize competitions for accurate labor market predictions. A new AI Workforce Research Hub would study disruption and policy options. These elements aim to address the human side of rapid technological change. They reflect input from the earlier House AI task force recommendations.
Supporters see the draft as a necessary step. A fragmented regulatory environment, they contend, slows progress and weakens American leadership. Industry groups have long lobbied for national standards. Companies such as OpenAI, Anthropic, and Google DeepMind already partner voluntarily with standards efforts. The bill seeks to formalize and expand that cooperation.
Critics counter that three years is too long to wait. State laws in California and elsewhere grew out of specific incidents and local priorities. Preempting them without stronger federal mandates, they say, creates a vacuum. The draft stops short of making federal vetting compulsory, a point that drew internal debate during negotiations.
House Speaker Mike Johnson signaled support for the preemption approach. It fits the administration’s light-touch philosophy. Yet House Democrats continue internal discussions on their AI stance. A separate Democratic commission on the topic serves as one forum for ideas.
The timing feels urgent. Lawmakers face an August recess and looming midterm elections. This draft represents what Obernolte called the last realistic window for Republicans to shape federal AI policy. Success is far from guaranteed. Bipartisan agreement on the core text could still fracture over details or amendments.
GAO reports and NIST pilot programs round out the package. One study would examine regulatory barriers to innovation. Another pilot focuses on better AI model documentation. International cooperation provisions appear in later titles, though specifics remain secondary to the domestic governance fight.
The release drew immediate attention on social media. Observers noted the constitutional questions around preemption authority. Others highlighted the section-by-section summary posted alongside the draft. That document, available on Obernolte’s congressional site, walks through each proposal in plain terms.
Whether this framework advances or stalls will shape the next phase of American AI policy. States have moved while Congress hesitated. The federal government now attempts to reassert control. The outcome will determine if innovation flourishes under one set of rules or if competing standards persist. And the debate over how much risk society should accept continues without clear resolution.
Additional recent coverage from Bipartisan Policy Center analysts underscores the difficulty of preempting without detailed federal standards in place. Their December 2025 analysis warned against acting first and legislating later. The current draft tries to avoid that trap by pairing preemption with new institutions and disclosure mandates. Even so, reactions suggest the balance remains delicate.
WebProNews is an iEntry Publication