5-Min Brief: Congress Just Dropped a 269-Page AI Bill. Here's What It Would Do to Every State AI Law in America.

What you need to know — in 30 seconds

• Two House members — one Republican, one Democrat — released a 269-page AI bill yesterday called the Great American Artificial Intelligence Act
• The most controversial provision: states would be banned from regulating how AI models are built for three years — freezing laws already passed in California, New York, and Illinois
• States could still regulate how AI is used after it's deployed — civil rights, consumer protection, workplace rules all remain
• The bill is a discussion draft, not a law — but it signals where Congress is heading on AI regulation

The United States has been watching AI regulation happen in slow motion. California passed laws. Colorado passed laws. Illinois passed laws. New York passed laws. Around fifty states have active AI legislation in various stages. The result is a patchwork of rules that AI companies say is impossible to navigate — and that consumer advocates say is the only protection Americans currently have.

Yesterday, Congress proposed a new answer to that problem. The question is whether it's a solution or a different kind of problem.

What the bill actually does

The Great American Artificial Intelligence Act — introduced as a discussion draft by Rep. Jay Obernolte (R-CA) and Rep. Lori Trahan (D-MA), with four other co-sponsors — is the most comprehensive federal AI legislation proposed so far. At 269 pages it covers a lot of ground. Here's what matters.

The three-year state preemption. The bill's most significant and contested provision is straightforward: for three years after the bill passes, no state can regulate how AI models are developed or built. The exact language: "No state or political subdivision thereof may establish, continue in effect, or enforce any law or regulation specifically regulating the development of any artificial intelligence model."

This would freeze California's AI training data transparency law, Colorado's AI consumer protection act, and similar laws in New York and Illinois. The preemption expires after three years — in theory giving Congress time to build a permanent federal framework before states can start again.

What states keep. The preemption doesn't apply to AI use — only to AI development. States can still regulate how AI is deployed in their borders. Civil rights protections, workplace rules, consumer privacy, child safety laws — all of those remain valid. A company's AI hiring tool can still be subject to state anti-discrimination law. What a company can't be regulated on is how it trains its model in the first place.

The federal framework it creates. The bill would establish a new AI governance structure with four pillars: frontier AI model oversight, workforce impact monitoring, cybersecurity standards, and AI research and development. It would fund a Center for AI Standards and Innovation at $100 million per year — the renamed version of the Biden-era AI Safety Institute. Frontier AI developers would be required to report critical safety incidents to the government and establish whistleblower protections for employees who raise concerns.

The frontier model transparency requirement. Large AI companies would be required to open their models up to government safety testing — a provision that directly targets OpenAI, Anthropic, Google, and Meta.

Who's for it and who's against it

The bill dropped yesterday and reactions came in fast.

Big Tech is largely supportive. Meta, Google, and OpenAI have all spent significant lobbying effort pushing for federal preemption of state laws. A single federal standard is easier to comply with than fifty different state regimes — even if that standard has more teeth than they'd like. OpenAI called the bill "a strong starting point."

Consumer advocates are furious. Public Citizen called it "a disastrous proposal that Big Tech is celebrating." Brad Carson, president of Americans for Responsible Innovation, said the bill "takes the current floor on state AI legislation and turns it into a federal ceiling." The argument: Congress has a long history of promising to build federal protections after blocking state protections — and then failing to follow through. If states are frozen and Congress doesn't act, Americans end up with no protection at all.

The Colorado AI Act is directly in the crosshairs. Colorado passed a landmark AI consumer protection law requiring companies to disclose when AI makes consequential decisions about people — hiring, lending, housing. That law, which has been held up as a model for other states, would be frozen under this bill.

The real fight underneath the bill

The technical debate about preemption is actually a proxy for a deeper disagreement about who gets to govern AI.

The AI industry's argument is simple: AI systems work across state lines. A model trained in California is used everywhere. Requiring it to comply with fifty different sets of training rules is impractical and slows innovation at a moment when the US is in a global race with China.

The counterargument is equally simple: state legislatures have historically been the place where consumer protections get built when Congress is too slow or too captured by industry interests. Data privacy is a perfect example — GDPR-style protections in the US have largely come from California, not Congress. Blocking states while Congress figures out AI governance is a bet that Congress will act decisively. That bet hasn't paid off on tech issues before.

Both arguments are reasonable. Neither is obviously right.

What it means right now

This bill is a discussion draft — it has not been formally introduced, and its sponsors explicitly invited public feedback. The email is even listed: GAAIA@mail.house.gov. Bills at this stage routinely get significantly amended or never pass at all.

But discussion drafts matter because they define the conversation. The specific provisions in this bill — the three-year preemption, the frontier model oversight framework, the safety incident reporting requirements — are now the reference point for every AI policy debate in Washington for the foreseeable future.

The US has been watching Europe regulate AI while struggling to pass anything at the federal level. Whether the Great American AI Act becomes law in anything close to its current form, it signals that the period of federal inaction may be ending. What replaces it will determine whether the US builds AI governance that protects people or governance that mainly protects the companies building AI.

HumanReadable-AI covers AI news in plain English every weekday. Subscribe below — free, no jargon, always under five minutes.