Washington wants one national AI rulebook, but states are already building the one companies will have to live with.
President Donald Trump warned states in December 2025 not to create their own artificial intelligence rules. Six months later, state lawmakers have mostly answered by moving faster.
According to reporting from the Associated Press, the White House has given no sign that it has followed through on the hardest parts of Trump's executive order, which directed the attorney general to challenge state AI laws and threatened some federal funding for states that pushed ahead. In the meantime, bills are moving through statehouses on child safety, employer use of AI, chatbot disclosures, synthetic media labels and the duties of developers building powerful models.
This is how regulation often arrives in the United States. Not as one clean federal statute, but as a stack of smaller obligations written by states, courts and agencies while Congress argues over the main event. For AI companies, that is a less dramatic story than a single national law. It may also be more expensive.
The first wave of AI bills tried to do too much at once. Governors and industry groups pushed back, and some broad measures died or were watered down. The new approach is more practical. Lawmakers are choosing the places where people already feel AI in daily life: a chatbot talking to a teenager, an algorithm screening a job applicant, a model creating a fake image, or a developer claiming its frontier system is safe enough to release.
Illinois is the clearest example. AP reported that legislation now on Democratic Gov. JB Pritzker's desk borrows from California and New York laws requiring large advanced AI developers to create protocols to prevent catastrophic harms, including biological weapons attacks, power outages and large-scale hacks. Illinois added an independent audit requirement, forcing developers to have an outside party review whether they are actually following their own safety policies.
That detail matters. A company can publish a safety policy with carefully chosen language and very little pain. An audit creates a record, and records are what regulators, plaintiffs and customers use when something goes wrong. The bill also drew nearly unanimous support, which is the kind of political fact that should worry companies hoping this fight stays neatly partisan.
Colorado has moved from another direction. Its AI law, signed in 2024 and taking effect on June 30, 2026 after delays, covers high-risk systems that substantially influence consequential decisions in areas such as employment, education, housing, banking, healthcare, insurance and legal services. AP reported that Colorado also passed a narrower May measure requiring companies deploying AI in those areas to tell people when the technology is being used to influence a decision about them.
For a worker, borrower or patient, disclosure is not a full remedy. But it changes the encounter. It tells the person that a machine is part of the process, and it gives regulators a cleaner trail if the system is later accused of discrimination.
Chatbots are becoming a child-safety issue
The most politically durable AI bills may not be the ones about model weights or national competitiveness. They may be the ones about children.
AP reported that Colorado, Connecticut, Idaho, Iowa, Nebraska and Oregon have passed chatbot-related laws this year, with several focused on minors. Connecticut's recent provisions apply to companion chatbots that sustain an ongoing relationship with a human. Those chatbots cannot interact with someone under 18 unless they are designed not to encourage self-destructive behavior and give parents tools to manage a child's use.
That is a hard issue for Washington to preempt cleanly. The White House has said it does not want to target state laws aimed at fraud, consumer protection and child safety. Its March 2026 national AI framework also urged Congress to address children, intellectual property and free speech while preempting state AI laws that conflict with its lighter regulatory view. The carveouts are where state lawmakers now have room to work.
Connecticut, Washington and Utah have also required AI developers to embed data into digital content so users can determine whether photos or videos were created or altered by AI. California lawmakers are advancing the No Robo Bosses Act of 2026, which would stop employers from relying only on AI to fire or discipline workers, and are also considering wider chatbot rules that include limits on using chatbot outputs from children for advertising.
None of this looks like one grand AI code. It looks like labor law, privacy law, consumer protection and child protection absorbing AI piece by piece.
The patchwork favors the companies that can afford it
The industry complaint about state AI laws is not baseless. A startup selling hiring software or a chatbot product across all 50 states may need lawyers, audits, data documentation, user notices and content-labeling systems before it has real revenue. A company such as OpenAI, Google, Microsoft or Meta can build compliance teams into the cost of doing business. A smaller developer may feel every new state rule as a new fixed cost.
That is the quiet moat forming under the politics. Large AI firms often say they want clear rules, and sometimes they do. But fragmented rules can also protect incumbents. If the cost of selling an AI product nationwide includes state-by-state legal review and independent safety documentation, the companies with the deepest balance sheets start each race several laps ahead.
Some Republican-led states have already hesitated. AP reported that Florida's House refused to advance Gov. Ron DeSantis' AI Bill of Rights after House Speaker Daniel Perez pointed to Trump's view that the federal government should lead. In Utah, a bill modeled on California and New York laws stalled after the White House warned lawmakers it opposed the measure.
Still, the larger direction is plain. Congress has not produced a federal AI law. The administration has not yet shown it can stop the states. Companies building or deploying AI are already facing a practical national regime, only it is being written in Springfield, Denver, Hartford and Sacramento before it reaches Washington.
Also read: Google is turning Gemini Omni into a video editing test for AI • Britain is turning teen safety into a tech compliance test • GM is keeping robotaxis alive by turning autonomy into a car feature
