Article 50 Enforcement: Who Can Complain, What Regulators Do, and the Letter You Should Be Able to Answer

Back to blog
Table of contents

General information, not legal advice.

Every regulation produces two kinds of fear: the informed kind, which changes behaviour, and the vague kind, which produces either paralysis or denial. Two months into Article 50, most of what fashion feels is the vague kind, because almost nobody has explained how enforcement mechanically works. This post does. The short version: the fine is real but distant, the complaint is cheap and near, and the difference between a bad month and a non-event is usually one page of preparation.

Who enforces, and who can start it

Enforcement of Article 50 is national: each member state's market surveillance authority applies the rules in its territory. 

The door into the process is wider than most compliance risks: any person can file a complaint. Not just consumers. A competitor who thinks your unlabelled synthetic campaign is buying an unfair advantage. A consumer association running a sweep of fashion e-commerce. An NGO with a thesis about synthetic bodies in advertising. A former employee. Complaints are free to file and require no proof of personal harm, which is what makes them, rather than inspections, the realistic trigger.

What actually happens, step by step

  • 1. The complaint. Someone files it, with screenshots. Note what they can see: your published images and any visible label, or its absence. They cannot see your internal records, which means complaints are almost always about the visible layer.
  • 2. Triage. The authority reviews whether the complaint is plausible and within scope. Weak complaints can die here; specific, well-documented ones proceed.
  • 3. The information request, the letter. The step that decides everything: a written request asking you to explain how your organisation complies. Which of your imagery is AI-generated, what marking it carries, how deepfake-class content is labelled, and who is responsible. There is a deadline. This letter is not an accusation; it is a test of whether an answer exists.
  • 4. What the answer does. A company that answers with a documented, coherent page has usually converted the case into a routine file. Gaps, contradictions or silence convert it into a proceeding, where orders and, eventually, fines of up to 15 million euros or 3 percent of worldwide turnover become available, alongside the publicity of being a named first case.

Proportionality runs through the whole process: authorities are expected to weigh cooperation, remediation and the gravity of the breach. Which is precisely why the letter stage matters so much: it is where cooperation is demonstrated or its absence recorded.

The one-page answer

Everything this blog has covered converges here. The letter asks four questions, and each maps to something a prepared team already has:

  • Which of your content is in scope: the inventory: which published imagery is AI-generated or AI-modified, by operation type (see which AI fashion images count as deepfakes and the EU AI Act Article 50 checklist).
  • How it is marked: the marking answer: what machine-readable layers your platform embeds, verifiable by anyone in a public checker, plus your export records as evidence of what left the building, when, with which marks.
  • How it is labelled: the labelling answer: the visible-label workflow for deepfake-class content, with the placement presets per channel.
  • Who is responsible: a name. The single most predictive fact in our survey data: organisations where someone owns AI-content compliance have answers; organisations where no one does have a scramble.

Write those four paragraphs now, calmly, and date the document. If a letter ever arrives, your first call is still to counsel, and what you hand them is this page instead of a panic. Our three-minute readiness check scores exactly these four dimensions, and the vendor half of the answers comes from the ten questions.

What enforcement is not

It is worth deflating two fears. There is no scanning van: authorities are not crawling fashion e-commerce with detection tools looking for unmarked pixels, and the rules explicitly do not require anyone to detect unmarked content, because the technology is not mature. And the first letter is not a fine: the path from complaint to penalty runs through the information request, and companies exit that path at the letter stage far more often than they proceed down it. The fine exists to make the letter impossible to ignore. The letter exists to find out who never prepared.

The regulatory calendar, duties and definitions live in the Article 50 guide, and the practical toolkit, from labels to records, is on the compliance page. If reading this produced a specific worry rather than a vague one, that is the informed kind of fear, and it comes with a to-do list, which is the entire difference.

Every image exported from Modelia carries signed provenance and an invisible watermark by default, and anyone can check any image, free, no account, in our public C2PA checker. The full picture, rules and industry data, is in The State of AI-Generated Content in Fashion 2026.

How would you rate this article