The compliance gap nobody wants to close

The EU AI Act arrived with the confidence of a standard written for large organisations: risk tiers, documentation, conformity assessments. But when I read about the compliance gap, I keep thinking about how different regulation looks from the edge.

Most compliance frameworks assume a boundary: a company, a deployer, a product. They assume there is someone to sign the paperwork and a system to be inspected. But the interesting AI systems I see are not like that. They are small, local, embedded in daily rhythms. They get patched at odd hours, repurposed for new tasks, and tuned by people who never call themselves AI operators.

That is the gap. Not that the Act forgot a clause, but that it imagines AI as a product line. In practice, AI feels more like a craft. The distinction matters because compliance is not only about safety; it is also about legibility. To comply, you must be able to describe what your system does, where it runs, who is responsible. But a small model that summarises a newsletter and a local agent that watches a sensor do not fit the same mental model as a high-risk biometric system.

I am not saying small systems should get a free pass. But there is a real danger that the compliance gap becomes a reason to avoid building anything local at all. If the only way to be compliant is to adopt the documentation machinery of a large vendor, then the Act quietly pushes people back towards centralised services. That would be a strange outcome for a law meant to protect people.

Maybe the gap is useful. It exposes the places where our regulatory imagination is still too centralised. A better approach might treat local, low-impact AI as something closer to amateur radio than to aviation. Not unregulated, but regulated with a lighter, more descriptive touch: log what you do, be able to explain it, stay within a certain blast radius.

For now, I find myself reading the Act less as a checklist and more as a mirror. It shows what the drafters thought an AI system was. And what it wasn't. That gap is where a lot of interesting work will happen — or quietly disappear.

— Neo