On 4 June 2026, Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA) released a discussion draft of the Great American AI Act (GAAIA), a bipartisan attempt to create a single federal framework for how the United States governs artificial intelligence. The important qualifier comes first: this is a discussion draft, not a bill that has been formally introduced. According to the lawmakers' own announcement, it was released specifically to gather feedback from stakeholders, experts and the public before any version is put before Congress — so what follows is a proposal under consultation, not law, and it may change substantially or not advance at all.

What the draft actually proposes

The draft runs to 269 pages. Legal and policy summaries describe it as spanning frontier-model oversight, workforce impacts, cybersecurity, federal AI adoption, and research and international cooperation. Its accountability provisions are aimed narrowly at the largest players rather than at AI broadly. DataGuidance and other legal summaries report that the heaviest obligations would apply to so-called large frontier developers — including companies above roughly US$500 million in annual revenue that have trained the most compute-intensive frontier models — which would leave most smaller developers and ordinary deployers outside those requirements.

On the institutional side, the draft would give statutory footing to the Center for AI Standards and Innovation, the Commerce Department standards centre that succeeded the US AI Safety Institute. According to FedScoop, it would authorise US$100 million per year for fiscal 2027 through 2029 for that body to develop voluntary guidelines and evaluate AI systems. In broad terms, the draft pairs accountability for the most powerful systems with a permanent federal home for standards-setting, which its sponsors argue the US currently lacks.

The central fight: one national standard versus fifty state laws

The most contested element is not the accountability rules but a proposed three-year preemption of certain state AI laws. The scope of that freeze matters, and is narrower than a blanket ban on state AI rules: according to FedScoop and other analyses, the draft would bar states from passing their own laws specifically regulating frontier-model development for three years, while still allowing states to regulate how AI is used and to apply laws of general applicability. Critics nonetheless see the development freeze as too broad. This is the genuine point of disagreement, and the draft's backers and critics split on it sharply.

Supporters frame preemption as a fix for fragmentation. In the sponsors' announcement, Representative Erin Houchin (R-IN) argued that a "patchwork of fifty different state laws" would make it harder for American companies to compete while doing little for consumers, and the sponsors' broader case — set out in an accompanying Bloomberg Law op-ed — is that AI risks do not stop at state lines, so protections should not depend on which state a person lives in. Representative Trahan, for her part, described the framework as establishing real accountability for the most powerful frontier systems without smothering innovation.

The opposition was immediate. Civil-society groups including Public Citizen, Public Knowledge and the AFL-CIO criticised the draft within days, and the House Democratic Commission on AI came out against it within hours of release; the preemption provision is the core of the objection. Critics worry that freezing state development rules would override or pre-empt protections states have already enacted or are moving to enact — Public Citizen, for instance, argued the bill strips states of authority to respond to real harms while leaving safeguards to a Congress that has yet to pass them. Industry groups such as the Business Software Alliance and the Information Technology Industry Council back the draft. In other words, both sides claim to be protecting consumers — supporters by ending the patchwork, opponents by preserving state floors — and the draft cannot fully satisfy both at once. That tension, rather than any single provision, is what will determine whether this becomes law.

Why this lands now

The draft did not appear in a vacuum. It came roughly two days after a White House executive order that set up voluntary federal agency reviews of new frontier AI models, and amid a broader federal effort to move from executive orders and voluntary reviews toward a more predictable national framework. That is the gap the draft is trying to fill: what statutory process should govern the most capable AI systems, and how much room should states retain while Washington builds that process? Whether this particular draft is the right answer is exactly what the consultation is meant to test.

The ASEAN read

For policymakers and AI businesses in Southeast Asia, the detail of US committee politics matters less than the direction. If the United States moves toward a single federal frontier-AI standard — with thresholds tied to model scale and revenue, and a designated standards body — it becomes another major reference regime alongside the EU AI Act, one that regional developers selling into the US, using US cloud or model platforms, or serving US-regulated customers would have to track. The compute-and-revenue threshold approach is also worth watching: defining obligations by model scale rather than by sector is a design choice ASEAN regulators weighing their own frameworks may borrow from or reject. And the preemption fight is a useful mirror for the region's own coordination problem — ASEAN faces a version of the same question across ten member states that the US is now arguing about across fifty, namely whether a common standard or a patchwork better serves both innovation and protection.

Key Takeaways

  • On 4 June 2026, Reps. Jay Obernolte (R-CA) and Lori Trahan (D-MA) released a discussion draft of the Great American AI Act — a 269-page bipartisan federal AI-governance framework. It is a draft for feedback, not a bill that has been introduced, and may change or not advance.

  • Its accountability rules are reported (per DataGuidance and other legal summaries) to target so-called large frontier developers — roughly companies above US$500M revenue with the most compute-intensive models — leaving most smaller developers and deployers outside the heaviest obligations.

  • It would codify the Center for AI Standards and Innovation (the former federal AI Safety Institute) in the Commerce Department, with US$100M/year authorised for FY2027–2029, per FedScoop.

  • The flashpoint is a proposed three-year preemption of state laws that specifically regulate frontier-model development (states could still regulate how AI is used): supporters call a state-by-state patchwork damaging, while critics — including Public Citizen and the House Democratic Commission on AI, which opposed it within hours — warn it would override existing state protections.

  • It landed about two days after a White House executive order that set up voluntary federal reviews of new frontier AI models, part of a broader push to move frontier-AI oversight from executive action toward a statutory framework.