The EU AI Act won't kill your AI program. Fog will.

In every European boardroom this year, the same scene: someone mentions an AI initiative, someone else says “but the AI Act,” and the room exhales with something that looks like relief.

Relief, because regulation is the perfect excuse. It’s external, it’s official, and it postpones the uncomfortable work of deciding anything. “Let’s wait for legal clarity” sounds like prudence. Often it’s paralysis with a compliance badge.

Let me say the quiet part: for most organizations, the AI Act is not what’s blocking you. I’ve worked inside European institutions and Belgian public administrations for a good part of my 26 years — environments with more regulatory constraint than almost any company will ever face — and I’ve watched genuinely ambitious digital work ship there anyway. The blocked organizations weren’t blocked by rules. They were blocked by not knowing what they wanted.

Compliance is a bounded problem

Here’s the distinction that matters. The AI Act — whatever its final calendar looks like as obligations phase in through 2026 and 2027 — is a bounded problem. It’s risk-based: it defines categories of AI use, from prohibited to high-risk to minimal-risk, and attaches obligations accordingly. Bounded problems are solvable with ordinary discipline: classify your use cases, document, assign accountability, monitor. Unpleasant, costly perhaps — but knowable. You can put it on a plan and retire it item by item. Legal and consulting firms will happily help; it’s exactly the kind of well-defined work they excel at.

Fog is a different species of problem. No direction, no owner of the outcome, no agreed picture of what AI should change in the organization. Fog is unbounded — you can’t scope it, schedule it, or outsource it. And no compliance program resolves it, because compliance answers “are we allowed to?” while fog is a failure to answer “what are we trying to do?”

Organizations routinely treat the second problem as if it were the first — commissioning legal reviews as a substitute for making choices. You can be perfectly compliant and perfectly stuck. Many will be.

What regulated environments taught me

Working at the European Parliament — tens of thousands of pages, 27 languages, every pixel subject to institutional scrutiny — taught me something counterintuitive: constraint is not the enemy of ambition; vagueness is. With a clear goal (in our case: measurable task success for over a million visitors a day), constraints become design parameters. Without one, even total freedom produces nothing but options.

The pattern repeats across every regulated sector I’ve worked in, from energy to public administration: teams that knew what they wanted treated regulation as a wall to build along. Teams that didn’t know what they wanted treated regulation as the reason they hadn’t built.

There’s even a strategic upside for those who move: surveys of AI leaders show that the organizations succeeding with AI implement governance before scaling — not as a brake, but as the thing that makes confident scaling possible. In 2026’s vocabulary: bounded autonomy, audit trails, escalation paths. Governance done right is what lets you say yes faster, with proof. For European organizations, mastering that early is a competitive asset, not a tax — the ones who wait for perfect legal certainty will be years behind the ones who built the muscle while regulating themselves.

The test for your organization

Ask your leadership team two questions, separately:

1. Which of our AI use cases fall into which AI Act risk category, and who owns bringing each into compliance?
2. What are the two or three workflows AI should transform here in the next 18 months, and who owns each outcome?

If question 1 has no answer, you have a compliance project. Manageable, hire help, execute.

If question 2 has no answer, you don’t have a regulation problem at all — and no lawyer will solve what you actually have.

And if question 1 keeps being discussed while question 2 keeps being postponed, name the dynamic honestly: the Act has become a socially acceptable place to store your paralysis. The companies that will own the European market in 2028 are answering both questions at once — this year.

Sources

• Official AI Act timeline
• Writer, Enterprise AI adoption 2026
• MachineLearningMastery, 7 Agentic AI Trends