Geopoliticsnorth

The Regulation Everyone Thought Would Unite Europe

Here’s something that genuinely excites me about democratic governance: When the European Union finally passed the AI Act in 2024, it represented something remarkable. For the first time, a major economic bloc was attempting to establish a comprehensive, proactive legal framework for artificial intelligence before the technology spiraled entirely beyond public accountability. Not after a crisis. Not in response to a catastrophe. Before. That’s the kind of foresight we should celebrate.

But here’s the thing about ambitious legislation in a 27-member union: the devil doesn’t just hide in the details. Sometimes the devil lives in the schedule itself. The EU AI Act’s phase-in timeline, which began when the regulation formally entered into force in August 2024, has created something unexpected and troubling. Instead of producing the unified European AI governance framework everyone envisioned, it’s actively generating regulatory fractures within the bloc. Understanding why this happened tells us something important about how democracies actually implement complex policy.

By February 2025, the Act’s prohibitions on unacceptable-risk AI systems became fully operational across all member states simultaneously. That sounds orderly on paper. In practice, it’s been anything but.

When National Interests Collide With European Rules

Let’s walk through what happened in late 2024. France and Germany, the EU’s two largest economies, jointly launched what they called a “strategic autonomy” lobbying effort. Their argument was straightforward: the AI Act’s regulatory framework, while well-intentioned, threatened to handicap European companies, particularly foundation model developers, relative to American and Chinese competitors operating in less restrictive environments. They weren’t wrong about the competitive reality. According to the Stanford HAI 2025 AI Index Report, Europe produced just 7 percent of the world’s notable AI models in 2024, compared to 61 percent from the United States and 15 percent from China. Those numbers matter politically. They matter to governments trying to defend industrial policy choices to their voters.

What makes this interesting, and what makes it a genuine governance problem, is that France and Germany weren’t asking for the Act to be dismantled. They were asking for carve-outs. Strategic exceptions. For foundation model developers, primarily. The logic was economically defensible. The precedent it set was less so.

When the two largest member states publicly lobby for exceptions to flagship EU legislation before that legislation is even fully implemented, other member states notice. They take notes. They start asking: if Germany and France get special treatment, what’s our carve-out going to be? Suddenly you’ve moved from having a unified regulatory framework to having a patchwork of national interests competing for exceptions. That’s not governance. That’s the erosion of governance in real time.

The Enforcement Capacity Crisis Nobody’s Talking About

Here’s where the structural problem becomes undeniable. The European Commission established a dedicated AI Office to oversee compliance with the Act. Reasonable people might assume an agency responsible for regulating one of the world’s most consequential technologies would receive adequate resources. What actually happened is worth examining. The AI Office received a staffing budget of approximately 9.7 million euros for 2025. That’s the sum total for a regulatory body tasked with monitoring AI compliance across 27 member states.

Members of the European Parliament’s IMCO committee (the Internal Market and Consumer Protection Committee) called it woefully insufficient. They were correct. You could parse the technical debate about what constitutes “adequate” funding, but the functional reality is simple: enforcement requires people. It requires investigators, legal specialists, data scientists who can actually audit AI systems. Nine-point-seven million euros doesn’t buy you a meaningful enforcement infrastructure across an entire continent.

What this creates is a vacuum. When you have ambitious rules but insufficient enforcement capacity, you don’t get compliance. You get a compliance lottery. Well-resourced member states, the ones with robust data protection agencies, can begin meaningful enforcement. Italy’s Garante, for instance, issued its first formal AI Act compliance investigation notices in early 2025, targeting three generative AI service providers operating in the Italian market. That’s real enforcement happening. Smaller member states without equivalent institutional capacity? They struggle. Companies quickly figure out where the pressure is and where it isn’t.

How Phase-In Schedules Become Political Weapons

What we’re watching unfold is a familiar pattern in regulatory implementation. The phase-in period, which was supposed to give everyone time to prepare, has instead created a window for strategic maneuvering. The European Commission EU AI Act Official Text and Timeline laid out a clear schedule. But schedules are political documents as much as technical ones. They create moments where different actors, member states, industry groups, civil society organizations, can make their case for modification.

The question now is whether the EU can stabilize around genuine enforcement, or whether it will fracture into a system where the AI Act exists on paper but operates differently depending on which member state you’re in. That’s not a rhetorical concern. We’ve seen it happen with other EU regulations, data protection rules that look identical in the official text but play out very differently depending on how aggressively individual national authorities choose to enforce them.

The encouraging part is that this fracturing isn’t inevitable. It’s visible. It’s documented. Members of the European Parliament, civil society organizations tracking this, journalists reporting on it — they see what’s happening. In a democratic system, visibility creates pressure for correction.

What This Means for the Rest of Us Watching From Outside

The EU AI Act’s implementation challenges matter far beyond Europe’s borders. Every jurisdiction looking at how to regulate AI, which is to say every jurisdiction on Earth, is watching this experiment. They’re asking: Can you actually implement comprehensive AI governance at scale? Or do competing interests and enforcement realities make it impossible?

The honest answer is still being written. What we know right now is that ambitious policy and adequate implementation capacity have to move together. You can’t have one without the other and expect success. You also can’t assume that putting rules on paper automatically produces compliance, especially when those rules threaten powerful economic interests and enforcement infrastructure is underfunded. Those are lessons worth learning carefully, because a lot of other democracies are about to make similar bets on their own AI governance frameworks.

The EU AI Act remains genuinely remarkable legislation. The question now is whether European member states will commit to making it work, or whether they’ll let it become another beautiful legal framework that functions very differently depending on where you are when you use it. What do you think the path forward should be? Have you noticed how these governance questions play out in your own local context? I’d love to hear how you’re thinking about this.