The Short Version
Two days after President Trump signed a voluntary AI review executive order, a bipartisan 269-page discussion draft — the Great American AI Act — landed in Congress on June 4, 2026. Its most explosive provision: a three-year freeze on any state law regulating the development of frontier AI models. The bill would override Colorado’s AI Act (effective June 30), California’s AB 2013 training-data disclosure rules, and New York’s RAISE Act. Frontier developers earning over $500M/year would be required to publish catastrophic risk frameworks, hire NIST-licensed auditors, and report incidents within 15 days — or face penalties of $1M per day.
🔍 THE BOTTOM LINE
The bill’s supporters say America needs a single national framework to compete with China. Its opponents call the preemption clause a “generational mistake” that locks out fifty-state consumer protections for three years — while Colorado’s law goes live in 25 days.
The Three-Year Freeze — What It Means
The core of the bill — and the reason it’s already drawing fire — is the preemption clause:
States would be barred from enacting or enforcing any law that “specifically regulates the development of an AI model” for three years from enactment.
What counts as “development”? Everything from model training and safety testing to developer accountability during the build phase. States could still regulate how AI is deployed and used — but anything touching the training or safety pipeline would be federal turf.
What Gets Preempted
- Colorado’s AI Act — Takes effect June 30, 2026 (25 days from the bill). Requires risk assessments and consumer protections for high-risk AI systems. The federal bill would override it entirely.
- California’s AB 2013 — Mandates AI developers disclose training data provenance. Preempted.
- California’s SB 942 — Requires watermarking of AI-produced content at the development stage. Preempted.
- New York’s RAISE Act — Sets safety requirements and testing for frontier models. Directly in the bill’s crosshairs.
Rep. Erin Houchin (R-IN), one of six sponsors, put it bluntly: “America should lead the world in artificial intelligence, not regulate ourselves into falling behind China through a patchwork of fifty different state laws.”
What Survives
The preemption explicitly does not apply to laws related to consumer protection, privacy, civil rights, employment discrimination, or content labelling — so state laws on AI use and deployment in those areas would remain intact.
What Frontier Developers Would Have To Do
Companies earning over $500M/year from AI model development or deployment face a significant compliance burden:
| Requirement | Detail |
|---|---|
| Risk Framework | Publish and maintain a catastrophic risk management framework covering biosecurity, cybersecurity, and societal harms |
| Independent Audit | Hire NIST-licensed auditors to assess compliance annually |
| Incident Reporting | Report any serious safety incidents to the new federal oversight body within 15 days |
| Penalty | $1M per day for non-compliance with safety provisions |
The bill also authorises $100M per year through 2029 for a new federal AI standards centre called CAISI (Center for AI Safety and Innovation).
The Political Battle Ahead
The bill faces three immediate obstacles:
-
California vs. The Feds — California’s legislators have already signalled they’ll fight preemption, arguing that Silicon Valley’s home state has the deepest understanding of AI risks and should regulate accordingly.
-
The Colorado Clock — With Colorado’s law going live on June 30, there’s a 25-day window where state and federal frameworks could conflict. Businesses operating in Colorado will be caught in the middle.
-
Safety Groups Mobilise — AI safety advocates and consumer protection organisations are calling the preemption a “free pass for frontier labs to avoid accountability for three years.” Expect lobbying battles on both sides as the bill moves through markup.
Why This Matters for New Zealand
The US federal approach will influence how other countries — including NZ — think about AI governance. If the US settles on a federal framework with strong enforcement and a standardised safety testing regime, it could set a benchmark that NZ’s Ministry for Regulation references in its own AI guidance (released late May 2026). Conversely, if the preemption clause triggers a prolonged legal fight between states and the federal government, it could reinforce arguments that NZ should move faster on its own regulatory framework rather than waiting for global norms to settle.
What’s Next
- June 30 — Colorado’s AI Act takes effect (unless preempted)
- Late June / July — Markup sessions in the House Research and Technology Subcommittee
- Unknown — Whether the bill gathers enough bipartisan support to reach a floor vote before the November midterm election window
The Great American AI Act is not law yet. But it’s the most significant federal AI regulation proposal to date — and the three-year preemption fight is just getting started.
Sources: Roll Call, Cybersecurity Dive, TechTimes, Awesome Agents, JD Supra
