Congress Drops 269-Page AI Bill That Would Override Every State AI Law for Three Years

Late Thursday, June 4, the most ambitious federal AI governance document in U.S. history landed without fanfare on Capitol Hill. Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA), flanked by four co-sponsors from both parties, unveiled a 269-page discussion draft they’re calling the Great American Artificial Intelligence Act. If it passes—and that remains a substantial if—every state AI law in the country affecting model development would be frozen for three years. Congress would own the field.

The bill’s release is a direct response to a fragmented regulatory landscape that has been accelerating. California has multiple AI laws on the books and more in the pipeline. Colorado’s AI Act—the country’s first comprehensive state AI law—takes effect on June 30, 2026, just 25 days away. New York, Illinois, and Texas have advanced bills at various stages. The AI industry has been lobbying for a federal standard that clears the field of this patchwork; the Obernolte-Trahan draft is the most concrete legislative attempt to deliver one.

What the Bill Actually Does

The headline provision is a three-year preemption of state laws and regulations that “specifically regulate the development” of AI models. The preemption window includes a sunset clause, meaning it would expire automatically without reauthorization. Critically, the bill’s preemption does not extend to AI use and deployment regulations—states retain authority over how AI systems are applied in employment, housing, healthcare, credit, and other domains.

In practice, specific California laws would fall: AB 2013, which requires AI developers to publish summaries of their training data, would be preempted. Portions of SB 942, which mandates watermarking of AI-generated content, would also be overridden at the development layer.

For frontier labs, companies with annual revenue exceeding $500 million, the bill creates a new compliance category:

• Must publish public Frontier AI Safety Frameworks documenting their approach to catastrophic risk
• Must disclose whether their systems carry potential for catastrophic outcomes: defined as 50 or more deaths, or $1 billion or more in property damage
• Mandatory engagement with Independent Verification Organizations for semi-annual third-party audits
• Incident reporting: critical safety incidents within 15 days of discovery; imminent catastrophic risks within 24 hours
• Penalties: up to $1 million per day for non-compliance

The bill would also formally codify the Commerce Department’s Center for AI Standards and Innovation—the AI Safety Institute established under Biden—and authorize $100 million annually from 2027 through 2029 for AI research, education, and standards development. Whistleblower protections for AI safety violations are included. Criminal penalties apply for using AI to impersonate government officials.

The Preemption Fight

The battle lines formed almost immediately. Brad Carson, president of Americans for Responsible Innovation and a former Democratic congressman from Oklahoma, called the preemption provision a “generational mistake.” His argument: the existing floor on state AI protections would become a ceiling, preventing states from acting on harms that Congress fails to address in time.

Public Citizen framed it more sharply, arguing the bill “strips states’ authority to protect consumers, workers, and children” under the cover of federal uniformity. The Alliance for Secure AI added that even accepting a federal framework, the current draft “falls short” in protecting Americans from advanced AI dangers.

Labor unions issued joint rejections. The characterization circulating in labor press is blunt: “Hard no. This bill is a giveaway to the AI industry.”

From the tech industry side, the reactions were warmer but mixed. ITI and NetChoice praised the framework’s structure. However, NetChoice flagged concerns about aggressive auditing provisions that could force companies to disclose proprietary trade secrets to third-party auditors—a process they argue opens the door to competitive intelligence gathering under regulatory cover.

Why Bipartisanship Here Is Unusual

Federal AI legislation has been stalled for years. The Senate AI Working Group produced a roadmap in 2024 that never materialized into legislation. Multiple House AI hearings produced recommendations but no bills. The Obernolte-Trahan draft stands out precisely because it has genuine bipartisan backing and substantive text, not just a framework memo.

Rep. Obernolte, a former software engineer who chairs the House AI Task Force, has spent two years studying the issue. Rep. Trahan, from the Democratic side, brings credibility with labor and consumer advocates—though her endorsement of preemption is straining those relationships. The co-sponsors span the ideological spectrum: Suhas Subramanyam (D-VA), Scott Franklin (R-FL), Scott Peters (D-CA), and Erin Houchin (R-IN).

The White House has not commented publicly on the draft. The Trump administration’s recent executive order on AI—signed the same week—focused on promoting American AI leadership internationally through a voluntary review framework, leaving domestic regulatory architecture to Congress. That the White House held its fire on this bill suggests it may not be opposed to the direction.

The Colorado Problem

Timing is everything here. Colorado’s AI Act, which imposes requirements on developers and deployers of high-risk AI systems covering employment, education, finance, healthcare, housing, and legal services, takes effect June 30, 2026. If the federal bill passes and becomes law before then, Colorado’s development-related provisions could be preempted immediately.

That scenario is unlikely given the congressional calendar—the bill is a discussion draft, not yet formally introduced, and would need committee hearings, markup, floor votes in both chambers, and presidential signature. No federal AI bill has cleared Congress in any form. But the pressure it creates is real: Colorado’s government is now watching a bill that could nullify its landmark law before its first anniversary.

What “Discussion Draft” Actually Means

A discussion draft is an official step below a formal bill introduction. It invites feedback from stakeholders, academics, industry groups, and civil society without triggering the formal legislative clock. It signals seriousness—this isn’t a press release—but also flexibility. Obernolte and Trahan are explicitly asking the public for comment before finalizing the text.

That means the preemption scope, the revenue threshold for frontier lab classification, the audit mechanics, and the penalty structure are all still negotiable. The three-year sunset, the most contentious element, may be the first thing to shift under pressure.

What is not negotiable, by any reading of the sponsors’ intent, is the core proposition: that a single national framework for AI development should replace the emerging patchwork of state laws. Whether Congress can actually deliver one, before that patchwork becomes entrenched, is the defining regulatory question of 2026.