On July 27, a regulation took effect in Brussels whose only real purpose was to postpone a different European law due to apply six days later.
The law being postponed is the AI Act, passed in 2024 and sold to the world as the first comprehensive rulebook for artificial intelligence. The rescue is Regulation (EU) 2026/1744, the Digital Omnibus on AI. It pushes the obligations for high-risk AI systems from Aug. 2, 2026, to Dec. 2, 2027, and for AI built into regulated products to August 2028.
The reason should interest every member of Congress who believes America needs a comprehensive AI statute of its own. Brussels did not lose its nerve on safety. It postponed the law because the harmonized technical standards companies were supposed to comply with were not finished, and because several member states had not appointed the regulators meant to enforce them. The deadline arrived before the definitions did.
I have run a small marketing company in Bulgaria since 2015. When the General Data Protection Regulation took effect in 2018, my company was 3 years old, and the compliance work landed on the same handful of people who were doing everything else. That is how these laws distribute their weight. A 20-person firm has no compliance department. The large platforms lawmakers actually had in mind do have one, they staff it up, and the rulebook becomes their moat.
The European Commission half-admits this now. It has set itself a target of cutting administrative burden by 25% for all businesses and 35% for small ones, a reduction of 37.5 billion euros in recurring administrative costs by 2029. You do not set that target unless you know what you built.
Congress returns this month to a version of the same argument. State legislatures have considered more than 1,500 AI bills this year, and dozens of those bills are now law. The discussion draft that Reps. Jay Obernolte and Lori Trahan released in June, the Great American Artificial Intelligence Act of 2026, runs 269 pages and would freeze state laws governing how AI models are built for three years while placing new federal obligations on frontier developers.
The impulse is right. A company selling software in five states should not have to satisfy five definitions of algorithmic discrimination. But the cure on offer resembles the disease. Europe’s experience does not show that the problem was the number of rulebooks. It shows what happens when you write any rulebook that sorts a technology into risk tiers and annexes years before anyone can say what compliance means in practice. Europe tried that with some of the best legal drafting talent in the world and had to reach for the brakes.
THE PENTAGON’S FREEZER MYSTERY HOLDS A WARNING ABOUT MILITARY AI
There is a conservative alternative that does not need 269 pages. Fraud is already illegal, and so is discrimination in hiring and lending, and so is deceptive advertising. None of those laws cares whether a person or a model did the deed. Enforce them against conduct, keep preemption narrow enough to stop states from dictating how models get built, and let evidence of real harm, rather than a taxonomy written in advance, decide what deserves a new statute.
Americans have watched Europe regulate technology for a decade and taken the wrong lesson from it. The lesson is not that Brussels hates business. Plenty of people there want European companies to win. The lesson is that a comprehensive statute freezes a snapshot of a technology that refuses to hold still, and the cost of that mistake falls hardest on the firms with nobody spare to absorb it.
Boris Dzhingarov is the founder and chief executive of ESBO Ltd, a digital PR company based in Plovdiv, Bulgaria.