State AI regulation is still moving ahead in the U.S., even as the White House tries to keep artificial intelligence policy from becoming a state-by-state compliance map.
The Associated Press reported Sunday that lawmakers in multiple states are continuing to pursue or implement artificial intelligence rules despite President Donald Trump’s effort to curb state-level regulation. The push is not one single AI law. It is a growing collection of narrower rules around employment tools, automated decisions, child safety, chatbot disclosures, synthetic media, and discrimination risk.
That makes the fight more practical than philosophical for companies using AI. A business that deploys automated screening, scoring, recommendation, chatbot, or content-generation systems may not be waiting for one national AI law. It may already be dealing with state rules that attach to hiring, housing, lending, health care, education, advertising, consumer notices, or interactions with minors.
The federal order has not frozen state action
The White House’s current position is clear. In a December 2025 executive order, Trump argued that state-by-state AI regulation creates a costly patchwork and directed federal agencies to challenge state AI laws that conflict with a “minimally burdensome national policy framework.” The order created an AI Litigation Task Force, called for a Commerce Department review of state AI laws, raised the possibility of funding limits tied to the BEAD broadband program, and directed federal agencies to examine whether grants could be conditioned on states not enforcing certain AI rules.
But the order is not the same thing as a comprehensive federal AI statute. It also leaves room for some state activity. Its legislative recommendation section specifically says a federal framework should not propose preempting otherwise lawful state laws relating to child safety protections, AI compute and data center infrastructure, state government procurement and use of AI, and other topics to be determined.
That carveout matters because many state AI laws are not trying to regulate frontier model design directly. They are aimed at where AI touches consumers, workers, students, patients, tenants, borrowers, voters, advertisers, and children.
Colorado shows how the rules are getting narrower
Colorado remains one of the best examples of how state AI regulation is changing under pressure. The state’s original 2024 AI law was built around “high-risk” AI systems and algorithmic discrimination. In May 2026, Colorado replaced it with SB26-189, the Automated Decision-Making Technology Act, a narrower framework focused on automated decision-making technology used in consequential decisions.
The new law defines automated decision-making technology as systems that process personal data and use computation to generate outputs such as predictions, recommendations, rankings, scores, or classifications. It applies when those systems materially influence consequential decisions involving education, employment, housing, lending, insurance, health care, essential government services, or public benefits.
Starting January 1, 2027, developers must provide deployers with technical documentation describing intended uses, categories of training data, known limitations, and appropriate use and human review instructions. Deployers must give consumers clear notice at the point of interaction and provide a plain-language description within 30 days after a covered automated system materially influences an adverse outcome. Consumers can request access to relevant personal data, correction of factually inaccurate data, and meaningful human review and reconsideration after an adverse decision.
Colorado’s shift is important because it moves the debate from broad AI governance language into operational obligations: documentation, notices, data retention, adverse-action explanations, correction rights, and human review. That is the kind of rule a product, legal, HR, lending, or compliance team has to turn into workflow.
Employment AI is already becoming a state issue
Hiring and employment tools are another pressure point. The Illinois Department of Human Rights said on June 2 that the state’s AI employment provisions require employers using artificial intelligence and automated decision-making systems in hiring and employment to provide transparency and ensure those systems do not result in discrimination against protected classes.
Illinois temporarily postponed a June 10 public hearing on proposed rules, saying it was reviewing the rulemaking and coordinating with other state agencies. That delay is not a retreat from the underlying issue. It shows how hard it is to translate AI employment law into workable notice, audit, and discrimination standards without either leaving workers exposed or burying employers in ambiguous obligations.
For companies, employment AI is one of the riskiest categories because tools that look like productivity software can become decision systems once they rank applicants, summarize interviews, recommend candidates, score performance, or influence promotion and compensation decisions. The technical question is not only what model is being used. It is what data enters the system, what output reaches a human decision-maker, and whether the employer can explain, review, and correct the result.
Chatbot safety is becoming its own category
California’s companion-chatbot law points to a different state-level track: AI systems that simulate relationships or emotional support, especially for minors. Senate Bill 243, signed in October 2025, requires operators of companion chatbots to implement safeguards around AI interactions and gives families a private right to pursue claims against noncompliant or negligent developers.
This kind of law is less about enterprise risk scoring and more about product behavior: how a chatbot presents itself, how it handles vulnerable users, whether it encourages harmful dependency, and what obligations apply when a minor expresses self-harm or distress. It also sits in the area the White House order is least likely to erase completely, because the order itself identifies child safety as a category that a future federal framework should not simply preempt.
What companies should watch now
The useful way to read the state AI fight is not as a binary question of regulation versus no regulation. The live question is which obligations attach to which use cases.
AI used for ordinary drafting or summarization may carry a different risk profile from AI that materially influences hiring, credit, housing, health care, insurance, education, public benefits, or child-facing companionship. A chatbot disclosure rule is different from a synthetic-media label. A developer documentation duty is different from an employer notice requirement. A human-review right after an adverse decision is different from a broad model safety audit.
That means AI governance programs need a state-law view of deployment, not just a list of approved models. Teams should know where their AI systems are used, what decisions they influence, whether personal data is involved, which states users or workers are in, what notices are shown, how outputs are logged, and who can review or reverse an automated recommendation.
The federal government may still try to preempt or challenge some state AI laws. Congress may eventually create a national framework. For now, the compliance work is already happening closer to the product surface: in applicant tracking systems, loan workflows, school tools, customer chatbots, online ads, state procurement systems, and consumer-facing AI features.
That is why the latest state activity matters. AI regulation in the U.S. is not waiting for one sweeping federal statute. It is arriving as a series of narrower rules aimed at specific harms, specific workflows, and specific moments when software starts to make or shape decisions about people.
