Congress Unveils Great American AI Act: America's First Federal AI Governance Blueprint

When Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA) released a 269-page discussion draft on June 4, they did something Washington has largely failed to do for three years: produce a serious, bipartisan, comprehensive attempt to govern artificial intelligence at the federal level. The Great American Artificial Intelligence Act of 2026 is not law — it’s not even formally introduced yet — but it is, by the assessment of most AI policy analysts, the most substantive piece of federal AI legislation ever put to paper.

The timing is not accidental. The country is caught in a crossfire of competing regulatory impulses: Illinois passed its own frontier AI safety bill, New York attempted a data center moratorium, Colorado rewrote its AI law before it ever took effect, and the White House issued competing executive orders on AI innovation. The GAAIA arrives as a federalizing move — an attempt to pull the patchwork of state rules under one federal ceiling before they calcify into a Balkanized compliance nightmare for the companies developing AI at scale.

What the Bill Actually Does

The draft is organized into four titles. Title I — Frontier AI Governance — is the most consequential, and it breaks into three operational sections.

Section 111 requires large frontier AI developers to publish a “frontier AI framework,” a public-facing document explaining how they identify, assess, and mitigate catastrophic risks from their models. Developers must also file transparency reports on an ongoing basis and submit critical safety incident reports to a designated federal authority within 72 hours of a significant incident — a provision modeled loosely on SEC breach disclosure requirements.

Section 112 creates a licensing system for Independent Verification Organizations (IVOs) — third-party auditors certified by the federal government to conduct semi-annual assessments of frontier AI developers. IVOs evaluate governance policies, real-time risk monitoring systems, and catastrophic risk mitigation measures. This is the bill’s most novel mechanism: rather than having government agencies audit AI labs directly (politically contentious and technically demanding), it delegates that function to licensed private organizations operating under federal oversight.

Section 113 extends whistleblower protections to employees who report violations of the Act, shielding them from retaliation. All three sections sunset after three years, giving Congress a built-in forcing function to revisit the rules as the technology evolves.

Title II addresses workforce impacts, mandating better federal data collection on AI’s labor market effects and requiring additional transparency disclosures when AI is a “substantial factor” in qualifying mass layoffs — a direct response to the wave of AI-driven restructurings at Meta, IBM, and others. Title III covers cybersecurity for AI systems. Title IV funds research, development, and international cooperation, including provisions to ensure U.S. engagement with allied AI governance frameworks.

Who It Targets

The bill applies to “large frontier developers,” defined as companies with annual revenues exceeding $500 million that have trained a frontier-class model. That definition is deliberately narrow — targeting Anthropic, OpenAI, Google DeepMind, Meta AI, and xAI while exempting the vast majority of AI startups and developers. The co-authors framed this as scaling obligations to capability and resources, though critics argue the threshold creates barriers to entry that entrench incumbents.

The State Preemption Fight

The most politically charged provision is the three-year preemption of state laws specifically regulating how frontier AI models are developed. States retain the right to regulate AI use within their borders — employment discrimination, consumer fraud, healthcare applications — but lose the power to impose their own development mandates on the training and deployment pipeline.

The architects of the bill argue that a single federal standard, however imperfect, is preferable to 50 different state regimes imposing conflicting red lines on a technology that operates across jurisdictional boundaries. “Policy for a technology this transformative can only be built to last if it’s written by both parties,” Obernolte and Trahan wrote in a joint Bloomberg Law op-ed.

But the preemption clause ignited immediate opposition. The co-chairs of the House Commission on AI and the Innovation Economy, including Representative Ted Lieu (D-CA), announced the draft “cannot serve as the basis for productive dialogue” — a stinging rebuke from within the Democratic caucus. Public Citizen called it a bill that “strips states’ authority to protect consumers, workers, and children.”

The preemption also collides directly with the political calculus of the White House’s own recent executive order on AI innovation, which carves out national security carve-outs that the GAAIA does not fully address. Administration officials have not publicly endorsed the draft.

Why It’s Different This Time

Congress has attempted AI legislation before. The AI in Government Act (2020), the Algorithmic Accountability Act (multiple iterations), and various transparency bills have been introduced and quietly died in committee. What makes the GAAIA different is the specificity of its mechanisms. The IVO licensing system, the incident reporting timelines, the whistleblower protections, the workforce disclosure triggers — these are the details that turn aspirational language into enforceable rules.

The bill incorporates language from 12 existing bipartisan measures, a deliberate strategy to build a broader coalition. Six additional members — Reps. Scott Franklin (R-FL), Suhas Subramanyam (D-VA), Erin Houchin (R-IN), and Scott Peters (D-CA) — signed on at launch.

The discussion draft phase is also significant. Rather than introducing a bill and daring Congress to oppose it, Obernolte and Trahan explicitly invited public comment, framing the 269 pages as a starting point for negotiation. That approach buys political space but also opens the door to industry lobbying that could dilute the strongest provisions.

The Bigger Context

The GAAIA lands in a moment of unusual urgency. The Fable 5 jailbreak and subsequent government export control order demonstrated, in real time, that frontier AI systems can trigger national security interventions that affect millions of users and billions in revenue. OpenAI’s S-1 filing, which landed the same week, put AI governance under investor scrutiny in ways that quarterly earnings calls never quite managed.

Goldman Sachs projects $7.6 trillion in cumulative AI infrastructure investment through 2031. The companies making those investments have an obvious interest in legal certainty. The GAAIA, whatever its flaws, offers something the current regulatory vacuum cannot: a national framework that at least defines the terrain.

Whether Congress can move quickly enough to make that framework real before the technology makes it obsolete is the central question. AI capabilities are advancing on a horizon that doubles every four months, according to Anthropic’s own published data. Three years of Title I provisions will expire before the next generation of frontier models fully matures.

The discussion period closes in July. Formal introduction is expected in the fall. Whether it passes the Senate — and survives the constitutional challenges to its preemption provisions — is a different story entirely.