The Rulebook Arrived, But Not Everyone’s Playing by It
Here’s what happened in February 2025: the European Union’s AI Act prohibition on unacceptable-risk AI systems went live. After months of building anticipation and regulatory theater, the moment actually arrived. Companies across Europe woke up to enforceable rules about what AI they could and could not deploy. On paper, it looked decisive. Historic, even. The world’s first comprehensive AI regulation was no longer theoretical.
But talk to regulators, compliance officers, and tech policy folks actually working inside member states right now, and you’ll hear something messier. The grand unified rulebook is fragmenting. What was supposed to be a coherent European standard is becoming a patchwork where Germany follows one interpretation, Italy another, and France? France is pushing hard for exceptions that undermine the whole thing.
This isn’t a bug in the system. It’s what happens when you write continent-spanning regulations ambitious enough to reshape an entire industry, then try to phase them in while the biggest economies simultaneously lobby for carve-outs. You end up with the regulatory equivalent of a road map where half the drivers are looking at different versions.
When France and Germany Decided the Rules Didn’t Quite Fit
Let’s talk about what France and Germany were doing in late 2024. Both nations, worried that their AI companies were losing ground to American and Chinese competitors, jointly pushed Brussels for what they called “strategic autonomy” exemptions. Specifically, they wanted foundation model developers to have more flexibility in how they complied with the AI Act. The argument was straightforward: Europe produces only about 7 percent of the world’s notable AI models, according to the Stanford HAI 2025 AI Index Report, while the United States generates 61 percent and China 15 percent. How can European companies compete if they’re bound by rules that American firms can navigate more loosely?
This isn’t abstract economic anxiety. This directly affects what gets built where, what risks get taken, and who bears responsibility when things go wrong. If French and German regulators start interpreting the rules more generously for their own companies while Italian and Spanish regulators enforce them strictly, you don’t have harmonized regulation anymore. You have competitive advantage disguised as compliance flexibility.
The political pressure made sense from their perspective. But it also exposed something that regulatory books rarely show you: the tension between doing the ambitious thing and doing it evenly. Europe wanted to lead on AI safety. It also wanted to win on AI business. Those goals started rubbing against each other almost immediately.
The Enforcement Problem Nobody Wanted to Talk About
Here’s where the practical details get uncomfortable. The European Commission’s newly created AI Office got a staffing budget of approximately 9.7 million euros for 2025. Members of the European Parliament’s Internal Market and Consumer Protection Committee looked at that number and called it what it was: woefully insufficient. You cannot effectively monitor, investigate, and enforce compliance across a 27-member bloc with that kind of budget when you’re talking about an industry moving at the speed of AI development.
Think about what enforcement actually requires. You need experts who understand machine learning. You need people who can audit training data, test model outputs, investigate complaints, and coordinate with national regulators. You need the bureaucratic machinery to pursue violations across borders where member states sometimes have competing interests. And you need to do it faster than the pace of innovation. That’s not possible on this budget.
So what happens instead? National regulators step in and fill the gap. Italy’s data protection authority, the Garante, already issued its first formal compliance investigation notices in early 2025, targeting three generative AI service providers. That’s Italy solving its local problem in its local way. Which is fine, except Italy’s interpretation of the rules might differ from Germany’s, which might differ from France’s.
You can see where this leads. Instead of one EU standard, you get 27 different enforcement regimes, each one interpreting the same regulation through the lens of local political pressure, local business interests, and local capacity.
What This Actually Means Where You Live
Let’s bring this down to something real. If you live in a city trying to use AI for housing allocation or welfare eligibility determinations, the framework your city operates under depends partly on your country’s interpretation of the EU AI Act. A “medium-risk” AI system in one country might be “high-risk” in another. A company might comply with one member state’s enforcement and face investigation in another.
For local governments rolling out AI systems, this creates genuine confusion. Do you design to the strictest interpretation you can find, gambling that you’ll be fine everywhere? Do you customize for each market? Do you decide the regulatory risk is too high and stay out of certain countries? These aren’t abstract questions. They shape what services your community can access, how quickly your local government can innovate, and whether European companies stay competitive or cede the market to international players who face less fragmented regulation.
Here’s what matters: regulatory harmonization only works if enforcement harmonizes too. Right now it’s not. The political economy of AI governance inside Europe is revealing the gap between what sounds good in Brussels and what actually happens on the ground.
Why This Moment Still Matters
The temptation is to see this as a failed project. Europe tried to lead on AI governance and immediately discovered that leading is hard. There are contradictions between competitiveness and caution, between member state autonomy and continental coordination, between moving fast and moving carefully.
But here’s the thing worth remembering: fractured governance is what you get when you don’t regulate at all, too. At least now there’s a rulebook. At least there’s an actual European Commission EU AI Act Official Text and Timeline instead of every company writing its own ethical guidelines. The phase-in problems we’re watching are real. But they’re also fixable.
What happens next depends on whether member states double down on fragmenting the rules or recommit to making enforcement work. It depends on whether the AI Office gets adequate funding. It depends on whether France and Germany decide that strategic autonomy means something other than “our companies follow different rules.” The EU AI Act’s fractures aren’t inevitable. They’re choices being made right now, by real people in regulatory offices and national governments who can choose differently tomorrow.
Want to understand how your own city or region is approaching AI governance in response? Start by asking your local government representatives what they’re actually doing to comply with and enforce these rules. The answers might surprise you, and the conversation is worth having.