Summary
On June 4, 2026, Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA) released a 269-page bipartisan discussion draft of the Great American AI Act — the first comprehensive federal framework for governing artificial intelligence proposed in the 119th Congress. The draft's most contested provision would preempt state laws specifically regulating AI model development for three years while leaving state authority over AI deployment intact. Large frontier developers — companies with annual gross revenues exceeding $500 million — would face mandatory semi-annual third-party audits, required disclosure of catastrophic risk assessments, and civil penalties of up to $1 million per day for violations. The discussion draft has not been formally introduced as a bill; it circulates for public comment before that step.
What Happened
Representatives Obernolte and Trahan released the draft at a House Energy and Commerce Committee event on June 4, 2026. A discussion draft is a pre-legislative document distributed for public comment before formal introduction; sponsors use it to gather industry and advocacy feedback on specific provisions. Four additional co-sponsors joined the release: Representatives Scott Franklin (R-FL), Suhas Subramanyam (D-VA), Erin Houchin (R-IN), and Scott Peters (D-CA). Obernolte chairs the Subcommittee on Innovation, Data, and Commerce and has led the House's AI policy work since 2023.
The draft's preemption provision is its most structurally significant element. Under the proposed language, states could no longer enact new laws specifically governing how AI models are developed for three years following enactment. The preemption would not apply to laws governing AI deployment or use within a state, preserving state authority over AI applications in employment, healthcare, housing, and other domains. Existing consumer protection, civil rights, and privacy laws would also remain intact. Among the provisions explicitly targeted by the three-year freeze: California's AB 2013 (training data transparency requirements), portions of California's SB 942 (content watermarking), and frontier AI safety laws enacted in California, New York, and Illinois since 2023. The three-year sunset period was presented as a deliberate compromise — earlier bipartisan discussions had included ten-year preemption language.
The bill would codify the Center for AI Standards and Innovation (CAISI) in statute. CAISI was established administratively within the Commerce Department in the Biden administration as the AI Safety Institute. Under the proposed legislation, CAISI would receive $100 million annually for fiscal years 2027 through 2029 plus licensing fees from the Independent Verification Organizations the center certifies; its statutory authorities would terminate three years after enactment unless renewed. CAISI's functions under the draft include developing AI security standards, assessing risks from advanced AI systems, coordinating with federal agencies and international counterparts, and overseeing compliance audits.
For companies with annual gross revenues above $500 million — the draft's definition of "large frontier developer" — the bill specifies four mandatory requirements: publication of a public frontier AI framework disclosing catastrophic risk assessments (defined as potential for death or serious injury to 50 or more people, or $1 billion or more in property damage); documented cybersecurity measures for model weights; critical incident reporting within defined timeframes; and retention of CAISI-licensed Independent Verification Organizations for semi-annual compliance audits. Civil penalties for violations can reach $1 million per day. Criminal penalties attach specifically to AI-assisted financial crimes and AI-powered impersonation of federal government officials.
A voluntary early-access framework would allow AI developers to provide federal agency access to frontier models 30 days before public release, with government safety assessments shared in return. The draft does not address open-weight models, compute-based thresholds, export controls, or a preemption carve-out for states that pass laws matching or exceeding the federal floor — each of which had appeared in prior congressional proposals.
Opposition emerged immediately on the preemption provision. Brad Carson of Americans for Responsible Innovation called the three-year freeze a "generational mistake," arguing it would convert state consumer protections into a federal ceiling and prevent state-level responses to AI harms that emerge before the federal framework could act. Brendan Steinhauser of the Alliance for Secure AI opposed preemption without a floor equivalence, arguing any national standard should protect consumers at least as well as the laws it displaces. Industry critics, represented by NetChoice, raised concerns about the auditing and data-sharing requirements potentially exposing proprietary model details.
Why It Matters
The Great American AI Act discussion draft is the first federal legislative text from either chamber of the 119th Congress to directly address the AI state preemption question at this level of specificity. The provision, if enacted, would displace the most developed state-level AI governance frameworks in the United States — California's body of law in particular, which has served as the primary testing ground for AI regulation since 2023. Whether a three-year preemption suspends harm-prevention without providing compensating federal protection, or creates space for a functioning national framework to develop, is the central dispute the draft does not yet resolve.
The timing matters. The discussion draft arrived two weeks after Illinois unanimously passed its own mandatory frontier AI safety audit requirement, and less than four weeks before Colorado's comprehensive AI law takes effect on June 30, 2026. Congress is attempting to preempt state-level activity at the moment that activity is accelerating. The political viability of the three-year compromise depends partly on whether states perceive the federal audit requirements as genuinely equivalent to the protections they have enacted — a question that industry lobbying, consumer advocacy, and state attorneys general will contest through the formal comment period.
As a discussion draft without a bill number, committee assignment, CBO score, or floor schedule, the Great American AI Act has not yet entered the formal legislative process. The distance between a serious bipartisan discussion draft and an enacted statute is substantial. What the release does establish is the legislative vocabulary — state preemption, CAISI codification, semi-annual audits, $500 million revenue threshold, catastrophic risk assessments — that will structure federal AI governance debate through the remainder of the 119th Congress. Whether that vocabulary survives industry lobbying, state government resistance, and Senate negotiation in recognizable form is not determinable from the discussion draft stage.
§ How to read the metadata
- Landmark
- Fundamentally alters the trajectory; 2–5 per year.
- Major
- Meaningfully shifts the landscape; 2–4 per month.
- Notable
- Worth documenting; significance can be upgraded later.
- Confidence
- High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
- Contestation
- Uncontested = no formal challenge. Contested = at least one challenge open. Superseded = replaced by a later entry. Unresolved = dispute still open.