Every State Is Writing Its Own AI Rulebook. New Jersey Should Write Just One

The United States is regulating artificial intelligence the hard way. Congress has enacted exactly one AI-specific statute, the TAKE IT DOWN Act on nonconsensual intimate deepfakes, and comprehensive federal legislation remains stalled. Into that vacuum the states have poured roughly 1,200 AI bills, with 38 states adopting more than 100 AI laws in 2025 alone. January 1, 2026 brought a wave of them into effect at once: California’s frontier model transparency act and its companion chatbot law, Illinois restrictions on discriminatory AI in employment and the Texas Responsible Artificial Intelligence Governance Act. Connecticut and New York followed with statutes of their own this spring. Every one of these laws draws its own definitional lines, and Washington has noticed: a December 2025 executive order directs a Department of Justice task force to challenge state AI laws deemed inconsistent with a minimally burdensome national framework.

Now it is New Jersey’s turn. In March the Assembly Science, Innovation and Technology Committee advanced a five-bill package reaching consumer disclosure, licensed professions, real estate advertising, election chatbots and AI companions, and in June the Senate introduced its own AI companion safety bill. Each bill, read alone, is a sensible response to a real problem. Read together, they import the national patchwork into a single state. New Jersey should resist that instinct and enact one comprehensive AI statute instead.

What Trenton Has on the Table

The strongest of the bills is A4730, which would require a business deploying generative AI to interact with consumers in trade or commerce to disclose, clearly and at the outset, that the customer is not dealing with a human. A violation would be an unlawful practice under the New Jersey Consumer Fraud Act, which means treble damages, attorney fee shifting and Attorney General enforcement. The rest follow the same instinct in narrower settings. A4732 would require AI companion products to remind users every three hours that they are talking to software, at $15,000 per violation, while the Senate’s S4474 would add affirmative safety duties for companion operators, including protections for minors, at $1,000 per violation capped at $500,000. A4731 would direct the Division of Consumer Affairs to write a model generative AI policy for every licensed profession. A4728 would police AI-doctored real estate listing photos, and A4729 would require disclosure when chatbots dispense election information. Transparency, accountability and protection of the vulnerable run through all of them. These are the right values.

The Patchwork Problem, Imported

The trouble is structural. The bills do not share a single definition of the technology they regulate. They carry different penalty schedules, a per-violation penalty here, a capped aggregate there, a $500 first-offense fine somewhere else. They route enforcement through different machinery, the Consumer Fraud Act in one bill, the Penalty Enforcement Law in another, professional board discipline in a third. Nor are the bills homegrown. A4732 and S4474 visibly track California’s SB 243, down to the recurring three-hour reminder, yet the details diverge: California pairs its law with a private right of action, while S4474 expressly declines to create one, and each state defines the regulated product differently. That is exactly how fifty near-identical laws become fifty separate compliance projects.

Now put yourself in the position of a seed-stage founder with one AI product. A customer service assistant with memory and a friendly personality could plausibly fall under A4730 as a consumer-facing generative AI tool, under A4732 or S4474 if its personalization meets the broad statutory definition of an AI companion, and under whatever rules a professional board eventually issues if licensed professionals use it with clients. Three overlapping disclosure regimes, three penalty structures and no single place to look up the answer. Large enterprises absorb that mapping exercise as a line item. Startups absorb it out of runway. A regulatory structure that is hardest on the smallest innovators is exactly backwards for a state trying to build an AI economy.

One Law Is the Better Answer

A single New Jersey AI act could fix this without weakening a single protection in the pending bills. One statute means one set of definitions, so a company knows once and for all whether its product is generative AI, an automated decision system or an AI companion. It means one disclosure standard that scales with risk rather than with the accident of which bill got a committee hearing. It means one enforcement home, so businesses can predict their exposure and regulators can build real expertise. And it is New Jersey’s best insurance in the coming federalism fight. The December executive order exempts child safety regulation from its preemption agenda, which shelters the companion bills, but a Consumer Fraud Act disclosure mandate like A4730 is precisely the kind of law a federal task force will scrutinize. A carefully calibrated statute is far harder to caricature as an innovation-killing burden than a sprawl of overlapping mandates.

The obvious rejoinder is Colorado, which passed the nation’s first comprehensive AI act and then repealed and replaced it before it ever took effect. But the lesson of Colorado is that overbroad, compliance-heavy drafting fails, not that comprehensive legislation fails. Comprehensive does not mean maximal. It means coherent: unified definitions, risk-based obligations, safe harbors for small developers and a single enforcement channel. The developer carve-out in S4474, which declines to hold a foundation model provider liable merely because a third party built a product on its model, shows Trenton’s drafters already understand this kind of calibration. They should apply it once, at scale, rather than five times in five bills. Consistency argues the same way: through the Next New Jersey Program, the state is spending hundreds of millions of dollars to attract AI investment, and it makes little sense to recruit AI companies with one hand while handing them a fragmenting rulebook with the other.

What Founders Should Do While Trenton Decides

None of this is a reason to wait. The Consumer Fraud Act already reaches misleading AI-generated marketing, deceptive automated interactions and inaccurate AI outputs, and if you sell nationally, the state patchwork is already your problem. So inventory every consumer-facing AI touchpoint now, build clear disclosure into the product before any statute requires it, benchmark against the strictest regime you touch, likely California’s, and test your product against the pending companion definitions even if you would never use that word in a pitch deck. Disclosure designed in early is cheap. Disclosure retrofitted under subpoena is not.

New Jersey’s legislators have correctly identified the problems, and they are in good company: so has nearly every statehouse in America. The values in this package deserve to become law. New Jersey has the chance to do what Congress has not and what most states have not, which is to enact them as one law.