The Patchwork

If you build an AI hiring tool in California, you must disclose it. Move the same tool to Texas, and a different rulebook — the Responsible Artificial Intelligence Governance Act — applies instead. Deploy it in Colorado, and as of this year, a narrower automated-decision law governs only if the tool "materially" influences a "consequential" decision, a phrase three different state agencies currently interpret three different ways. Welcome to AI regulation in the United States in 2026: fifty different answers to the same question, and a Congress that just proposed making forty-nine of them illegal.

According to the legislative tracker Plural Policy, 19 new AI-related laws were enacted across 11 states and the federal government in a single two-week window ending in late June 2026 — a pace that has made "patchwork" the industry's least favorite, and most accurate, word. Watchdog groups now count more than 2,100 AI bills introduced nationwide this year alone. Some regulate hiring algorithms. Some regulate chatbot companions. Some regulate algorithmic price-setting at the grocery store. Almost none of them agree with each other on definitions, thresholds, or enforcement.

Three Laws, Three Philosophies

Start with the states that moved first. California's Transparency in Frontier AI Act (TFAIA) took effect January 1, 2026, requiring the largest AI developers to publish safety frameworks and report critical incidents — a disclosure-first model aimed squarely at frontier labs. Texas took a different route: its Responsible Artificial Intelligence Governance Act (TRAIGA), effective the same day, focuses on prohibited uses — AI that manipulates behavior to cause harm, or discriminates unlawfully — while otherwise staying out of how models are built. Illinois picked a narrower target still, requiring employers to notify job candidates when AI analyzes their video interviews, effective February 2026.

Then, in May 2026, Colorado did something unusual: it repealed its own AI Act — a broad 2024 law that had become a template other states were copying — and replaced it with SB 26-189, a narrower statute regulating only automated decision-making technology that "materially" influences "consequential" decisions such as housing, credit, or employment. The law doesn't take effect until January 2027, but its retreat from a broader framework is being read across the industry as the first sign that the most sweeping state AI laws may not survive contact with implementation.

The Federal Counter-Move

Into this scramble stepped a bipartisan group in Congress with the Great American AI Act, a draft framework unveiled in early June 2026. Its central provision: a three-year federal preemption of any state law that "specifically regulates the development" of an AI model. It follows a December 2025 executive order, "Eliminating State Law Obstruction of National Artificial Intelligence Policy," that had already signaled the White House's preference for a single national standard over what officials describe as an unworkable maze.

The bill's authors were careful to draw a narrower line than that rhetoric suggests. The preemption applies only to laws governing how AI models are built — not to how they're used. States retain full authority over AI in hiring, housing, healthcare, and financial decisions; consumer protection, privacy, and anti-discrimination law are explicitly untouched. In practice, that means California's TFAIA-style disclosure rules for frontier model developers would likely be frozen, while Illinois's interview-notification law and Colorado's automated-decision statute would survive largely intact.

Why the Fight Is Really About Speed

The deeper argument underneath the preemption fight isn't philosophical, it's operational. AI companies argue that complying with fifty overlapping and sometimes contradictory disclosure, audit, and notice regimes slows deployment and effectively hands an advantage to competitors in countries with a single national rulebook — including China's MIIT/SASAC-directed approach and the EU's harmonized (if slow-moving) AI Act. State legislators counter that Washington has had years to pass comprehensive AI legislation and hasn't, and that in the absence of federal action, states are the only bodies actually protecting people from algorithmic hiring discrimination, deceptive chatbots, and opaque pricing today, not eventually.

The Algorithmic Pricing Flashpoint

One of the fastest-moving fronts has nothing to do with frontier models at all: algorithmic pricing. California's AB 325, New York's S.7882, and Connecticut's HB 8002 all restrict companies from using AI systems to set individualized prices based on a consumer's personal data — the practice of charging you more because a model has inferred you'll pay it. Dozens of additional states are considering similar bills. Because this regulates use, not model development, it would survive the Great American AI Act's preemption language entirely — a preview of which fights are likely to be settled quickly and which will drag on.

Compliance costs from this patchwork also fall unevenly. A frontier lab with a dedicated policy team can afford to track nineteen new laws in a two-week window and adjust disclosure practices state by state; a five-person startup building a hiring-screening tool cannot, and often defaults to either over-complying everywhere (raising costs and slowing product launches) or under-complying somewhere (accepting legal risk it may not fully understand). Several state bar associations and small-business advocacy groups have flagged this asymmetry as a quiet consequence of the patchwork that gets less attention than the headline fight over frontier-model oversight: the practical effect of fifty overlapping rulebooks may be to entrench the largest AI companies, who can afford the compliance overhead, at the expense of smaller competitors who cannot.

An Old Fight in a New Wrapper

Washington has run this play before. Telecommunications and internet policy spent the better part of two decades cycling through nearly identical arguments over net neutrality: state legislatures in California and elsewhere passed their own open-internet rules after federal rules were rolled back, industry sued to have them preempted on the grounds that a patchwork of state internet regulation was unworkable, and courts spent years sorting out which slice of the issue was properly a matter of interstate commerce. That fight never fully resolved; it mostly exhausted itself as the underlying technology and business models moved on. AI lawyers on both sides of the current debate cite that history as a cautionary tale, though they draw opposite lessons from it — industry sees a decade of costly uncertainty worth avoiding this time by moving fast on federal preemption, while state legislators see two decades of proof that waiting for Washington to act first mostly just means waiting.

What's different this time is the compressed timeline. Net neutrality played out over roughly fifteen years. The AI patchwork has produced more than 2,100 bills in a single legislative session, and the federal preemption response arrived within six months of the first wave of state laws taking effect — a sign that everyone involved has learned from how slowly the last version of this fight moved, even if they haven't agreed on how it should end.

What Happens Next

Nothing about this resolves quickly. Even if the Great American AI Act passes in something close to its current form, its three-year sunset means the underlying disagreement — should AI be governed by fifty experiments or one federal standard — simply reconvenes in 2029, likely with more capable models, more incidents on the record, and more state legislators who ran for office specifically to close whatever gaps this compromise left open. In the meantime, an AI company operating nationally in 2026 needs a compliance map that updates roughly every two weeks, a genuinely new category of regulatory overhead that didn't exist three years ago.

For everyone else — job applicants, patients, shoppers whose prices might be quietly personalized — the practical answer to "what are my rights against an AI decision" now depends on which of fifty jurisdictions they happen to be standing in when they ask it.

That is likely to remain true even if the Great American AI Act passes this year. Its preemption window only covers laws about how models are built; the much larger and faster-growing body of state law governing how AI is used in housing, hiring, healthcare, and pricing decisions stays fully in force, and new versions of it are still arriving at a rate of roughly one per state per week. The patchwork isn't a temporary condition on the way to a single national rulebook. For the foreseeable future, it is the rulebook.

Sources