Washington wants one AI rulebook for America, while Brussels delays parts of its own

On December 11 President Trump signed an executive order titled Ensuring a National Policy Framework for Artificial Intelligence. Its purpose is to stop US states from writing their own AI rules. The Justice Department gets a task force to challenge state laws in court, states with rules the administration dislikes risk losing federal broadband money, and federal regulators are told to look for ways to override them. It lands three weeks after the European Commission proposed delaying the strictest parts of the AI Act by more than a year. For a European company that sells software on both continents, the two moves look similar from a distance and mean quite different things up close.

What the order does

According to Latham & Watkins’ summary, the order has five working parts. The Attorney General sets up an AI Litigation Task Force to challenge state AI laws the administration considers unconstitutional. The Commerce Secretary has 90 days to publish a list of state laws that conflict with a “minimally burdensome” national framework. States on that list lose access to part of their funding from the federal broadband programme, and agencies are asked whether other grants can come with the same condition. The Federal Trade Commission is to say when state laws that force AI models to change their outputs count as deceptive practices, and the Federal Communications Commission is to consider a federal reporting standard that would override state ones. Finally, the White House is to draft a federal law that preempts the states.

Some things are exempt: state laws on child safety, on data center infrastructure, and on how states buy AI for their own use. Colorado’s AI Act, which regulates algorithmic discrimination in areas such as hiring and lending, is singled out as the kind of law the administration wants gone.

An order cannot repeal a law

The weakness is legal. An executive order binds federal agencies. It does not strike down state law, which only Congress or the courts can do, and Congress has not done it. A proposal to put a moratorium on state AI laws into federal legislation recently failed. The order is an attempt to get by pressure and litigation what the administration could not get by vote.

The states are not backing down. Stateline reports that 280 state lawmakers from both parties signed a letter against federal preemption, and that 38 states adopted AI measures this year, with more bills already filed for 2026. Colorado state representative Brianna Titone called the order “hot air”. California and other states are reported to be weighing lawsuits. Republican states are among those passing AI laws, mostly on consumer protection and deepfakes, which makes this harder to frame as a partisan fight.

My expectation is years of litigation, with the state laws mostly in force while it runs. The order’s real effect is to raise the political cost of passing new ones.

Europe is loosening too, but differently

The Commission’s Digital Omnibus, presented on November 19, proposes pushing the AI Act’s rules for high-risk systems, such as AI used to screen CVs, decide on loans or grade exams, back to December 2027. They were due in August 2026. The Commission’s reasons are practical: member states are late in naming the authorities that enforce the rules, and the technical standards companies are supposed to follow are not ready. Consumer groups are unconvinced. Agustín Reyna of the European consumer organisation BEUC called the package deregulation almost to the exclusive benefit of Big Tech.

The difference is what remains afterwards. If the omnibus passes as proposed, Europe still has one AI law for 27 countries, with a later start date for one part of it. The obligations for general-purpose AI models and the bans on prohibited practices already apply. The US order goes the other way: it tries to remove fifty sets of state rules and offers, for now, no federal law in their place. Washington says it wants one rulebook. At the moment it has none.

The omnibus is also only a proposal. It needs the European Parliament and the member states, and the changes it makes to the GDPR in the same package are controversial enough to slow it down.

What this means for a European company

If you sell AI products or AI-supported services in the US, keep complying with the state laws that apply to you. Nothing in the order changes them today, and a law that is challenged is still a law until a court says otherwise. Colorado’s rules for high-risk AI decisions and California’s rules for frontier model developers are the ones to watch.

Do not pause AI Act preparation because of the omnibus. The delay affects the high-risk category, and only if it is adopted in time. The rules on AI literacy and prohibited uses already apply, and the transparency rules, such as labelling AI-generated content and telling people they are talking to a chatbot, are still set for August 2026. A company that uses the extra time to finish the job properly will be in better shape than one that stops.

And plan for divergence. The EU is heading toward one set of rules with a slower timetable. The US is heading toward a legal fight over who gets to write the rules at all. If your AI governance is built on the stricter European framework, you will cover most of what any US state asks for, whichever side wins in Washington.