America's 1,200 AI Bills and No Good Test for Any of Them: Inside the US Regulatory Impasse

The United States has introduced more than 1,200 artificial intelligence bills across Congress and state legislatures since 2023. According to a Fortune investigation published this month, there is no agreed-upon test for any of them — no shared definition of what “safe AI” looks like, no technical benchmark that legislators can point to, and no institution with the mandate and resources to evaluate whether a proposed requirement would actually reduce harm or simply add compliance paperwork.

That vacuum is now pulling in three competing forces simultaneously: a White House pushing for federal preemption of state laws, a Senate bill proposing the most comprehensive AI regulation in American history, and a Congress that has already rejected federal preemption by a vote of 99-to-1. Something will give. The question is what — and how fast.

The Scale of the Problem

To understand how the US arrived at 1,200 bills and counting, it helps to understand why the legislative pace accelerated so sharply. The release of ChatGPT in late 2022 sent lawmakers at every level scrambling to respond to constituent anxieties about AI-generated disinformation, job displacement, algorithmic bias, and child safety online. Without federal leadership, states moved first. California, New York, Texas, and dozens of others introduced or passed AI-related legislation covering everything from algorithmic hiring tools to deepfake pornography to autonomous vehicle liability.

By the time Congress woke up to the scale of the state-level activity, the landscape had fractured. Companies developing AI for national markets now face overlapping, sometimes contradictory requirements across multiple jurisdictions. A single AI-powered hiring tool might need to comply with Illinois’s algorithmic bias rules, New York City’s AEDT audit requirements, California’s automated decision-making regulations, and the requirements of any other state where the employer operates. Legal teams and compliance officers are overwhelmed; small AI companies, which lack the resources to track 50 state legislatures simultaneously, are beginning to concentrate their products in fewer jurisdictions.

“We’ve created a situation where the most motivated actors to shape AI governance are the ones with the most resources to do so,” one AI policy analyst told Fortune. “That’s not a coincidence.”

The White House Approach: Light Touch, Heavy Preemption

On March 20, 2026, the Trump administration published its National Policy Framework for Artificial Intelligence — a 40-page document outlining the White House’s preferred approach to AI governance. The framework calls for a “light-touch” regulatory philosophy grounded in existing agency authority, avoiding new bureaucracies or novel legal mandates. More controversially, it explicitly calls for broad federal preemption of existing state AI laws.

The administration’s position is that the patchwork of state regulations creates an unacceptable drag on American AI competitiveness, particularly relative to China. Trump’s December 2025 executive order had already directed the Department of Justice to challenge state AI laws it viewed as obstructing national AI development policy and conditioned certain broadband funding on states adopting a “minimally burdensome” regulatory standard.

The argument has economic logic. Regulatory fragmentation imposes real costs, and those costs fall disproportionately on the smaller AI companies that lack lobbying power in every state capital. A national standard — even an imperfect one — reduces that burden.

But the White House framework faces a fundamental political problem: Congress has already voted, loudly, against broad federal preemption.

The 99-1 Rejection

When the Trump administration’s “One Big Beautiful Bill” — the sweeping legislative package that combined tax cuts, immigration reform, and various domestic policy priorities — worked its way through the Senate in mid-2025, it included a provision that would have imposed a ten-year moratorium on state AI regulation. The provision was initially positioned as a compromise between a full preemption ban and the status quo. As it reached the Senate floor in its final form, it had narrowed to conditioning federal broadband funding on states’ agreement to suspend their AI enforcement.

Even this reduced version failed spectacularly. Senators Marsha Blackburn, a Tennessee Republican, and Maria Cantwell, a Washington Democrat, jointly introduced an amendment to strip the provision. The amendment passed 99-to-1, a margin that reflected not ideological agreement on what AI policy should look like, but near-universal congressional resistance to the executive branch dictating to states on their own consumer protection authority.

The final bill signed by President Trump on July 4, 2025, contained no restrictions on state AI legislation whatsoever.

Blackburn’s Comprehensive Alternative

Senator Blackburn drew the obvious lesson from the 99-1 vote: preemption through executive action or legislative sleight-of-hand was politically dead. She spent the subsequent months drafting an alternative — a comprehensive federal bill that would create a national framework robust enough that states might voluntarily defer to it, rather than being forced.

The result, released as a discussion draft in March 2026, is the TRUMP AMERICA AI Act — formally titled “The Republic Unifying Meritocratic Performance Advancing Machine Intelligence by Eliminating Regulatory Interstate Chaos Across American Industry Act.” The acronym is deliberate. The bill runs to 291 pages and is organized around what Blackburn calls the “4 Cs”: children, creators, conservatives, and communities.

Key provisions include a general duty of care for AI chatbot developers, requiring “reasonable care in the design, development, and operation” of AI systems to prevent foreseeable harms to users. The bill would require AI developers to regularly assess the risks of advanced models and report their safety protocols to the Department of Homeland Security — a reporting requirement that would apply to the frontier AI labs that have voluntarily submitted to government testing through the Coalition for AI Safety and Innovation (CAISI) framework, but would now impose mandatory obligations.

Most controversially, the bill would repeal Section 230 of the Communications Decency Act two years after enactment — eliminating the foundational legal protection that has shielded internet platforms from liability for third-party content since 1996. Combined with new AI-specific liability exposure for foreseeable harms, this provision has generated fierce opposition from tech industry groups.

On preemption, Blackburn threaded a needle: the bill expressly states it does not preempt generally applicable state laws, but specific titles — particularly those governing children’s safety online — do preempt conflicting state law while permitting states to set higher protective standards for minors. It is a more limited preemption than the White House wants, but enough to create a national floor.

“I think this has real momentum,” Blackburn told reporters after releasing the draft. The bill has attracted attention, though formal co-sponsors have moved cautiously in a chamber where AI policy remains deeply contested.

States Are Not Waiting

While federal negotiations proceed, states have continued legislating. California’s SB 53, enacted in 2025, requires frontier AI developers to publish safety protocols and conduct red-team testing before deploying new models above defined capability thresholds. New York’s Responsible AI Safety and Education (RAISE) Act mandates incident reporting when AI systems cause significant harm and created a new state AI oversight office. Texas’s Responsible Artificial Intelligence Governance Act (TRAIGA) prohibits specific intentional misuses of AI including voter suppression and discriminatory denial of essential services.

Colorado’s SB 189, which was rewritten substantially in May 2026 after the original AI Act sparked industry backlash, now focuses on high-risk AI systems in employment and credit with a risk-tiered approach that draws from the EU AI Act’s taxonomy. Colorado has become something of a laboratory for what state-level AI regulation can realistically achieve — and the SB 189 revision illustrates how difficult it is to balance consumer protection against developer compliance costs.

EU comparison remains instructive. The EU AI Act enters full applicability in August 2026, giving Europe a unified framework that, whatever its implementation challenges, allows companies to comply once across 27 member states. American companies operating in Europe must comply with the EU AI Act regardless of what the US does. For global AI developers, the EU’s requirements are increasingly becoming the de facto baseline, with US-specific rules layered on top rather than replacing them.

What Happens Next

The TRUMP AMERICA AI Act is not expected to pass in its current form. Its repeal of Section 230, its duty-of-care standard, and its DHS reporting requirements each face significant opposition from different parts of the coalition that would need to assemble to pass a 291-page bill through a divided Congress. Blackburn’s own statement frames it explicitly as a “discussion draft” — a starting point for negotiation rather than a ready-for-floor bill.

But discussion drafts shape the negotiation. The key questions that will determine what federal AI legislation, if any, passes in the current Congress are: Can a duty-of-care standard be scoped narrowly enough to survive tech industry opposition while retaining enough teeth to satisfy consumer advocates? Will Section 230 reform be packaged with AI regulation, or split into separate legislation? And how much preemption of state law is politically achievable after the 99-to-1 rebuke?

For AI developers and deployers, the practical implication is that the US regulatory environment will remain in flux through at least the end of 2026, with state laws continuing to proliferate. Tracking more than 1,200 bills across 50 state legislatures and Congress is not a problem that gets solved by ignoring it. Companies that build compliance infrastructure now — even for requirements that may be superseded by eventual federal legislation — will be better positioned than those waiting for regulatory clarity that may be years away.

The 1,200-bill count is not a stable equilibrium. It is the precursor to either consolidation around a federal framework or further fragmentation into a genuinely unworkable patchwork. Which outcome arrives first depends largely on whether Congress can agree on something — anything — before the next wave of elections reshuffles the legislative math again.