Discover stores
14 October 2026

The EU AI Act and what it means for fashion brands

Made with AI? The new EU rules every fashion brand needs to know.

Production
BLOG AI act cosh 3

The EU AI Act is Europe’s new law on artificial intelligence, and parts of it now apply to every fashion brand that uses AI. For most everyday uses, the core obligation is transparency, and it comes down to one simple idea: if AI made it and people could mistake it for real, tell them.

The AI Act sorts AI into risk levels. Most of what fashion brands do with AI, like campaign images, product copy, chatbots and outfit recommendations, falls into a group with light rules: you can use it, as long as you’re open about it. Those transparency rules started on 2 August 2026.

Some uses of AI count as ​“high-risk” and come with much stricter rules, and fashion brands can use them too. Examples include AI that helps recruit, select or monitor staff, and systems that recognise emotions, such as some in-store camera analytics. Most of these stricter rules start in December 2027. If you use AI in HR, in your stores or in any other area beyond marketing, check whether it falls into this category.

It doesn’t matter where your brand is based. If your content reaches people in the EU, and anything posted openly online does, the rules apply. Breaking them can be expensive: fines can reach €15 million or 3% of global turnover, with lower caps for smaller companies. But for most brands, the bigger risk is losing customers’ trust.

Who’s responsible: you, your AI supplier or your agency?

The law gives different jobs to different players.

Most brands are ​“deployers”. You use AI tools made by someone else, like an image generator or ChatGPT, and you decide what to use them for. Your main job is labelling, which we cover below.

Some brands become ​“providers”. If you train an AI tool on your own photo archive and use it under your own name, you may count as the maker of that tool. Then you also have to make sure its output is marked as AI-generated in a machine-readable way, such as a hidden watermark or metadata, and that it can be detected.

And your agency or freelance marketeer? It depends on who decides. If you simply order a campaign and the agency or freelancer chooses whether to use AI, the agency is responsible for the labels. If you tell them to use AI or control how they use it, the responsibility is yours. Write down who decides and who labels in your briefs and contracts.

Campaign images, virtual models and product photos

BLOG AI act cosh 2

This is where fashion is most affected. The law calls realistic AI images, videos or audio that could pass as real ​“deep fakes”, and they need a visible label.

  • A photorealistic AI model who doesn’t exist still needs a label, because she looks like someone who could exist.
  • An AI image that makes a jacket look softer, brighter or better made than it really is needs a label, and could mislead customers anyway.
  • If you use AI for editing purposes, for example background removal or colour correction, you do not need to label it.

The label has to be visible, for example ​“Created with AI” in the caption or on the image. The invisible watermark that some AI tools add doesn’t count. You can use Instagram’s ​“AI info” label, but making sure it’s there is still your job.

Not sure? Give your team one simple rule: realistic AI people or products get a label. When in doubt, label.

The good news: images and videos made before 2 August 2026 don’t need a label afterwards.

Sustainability claims and other text

If AI helps you write a product description, you don’t need a label, unless it makes claims about health, safety or sustainability.

That exception matters for our industry. Claims like ​“made with recycled fibres”, or articles about your climate impact, count as information the public needs to be able to trust. If AI wrote or rewrote them, you have two options:

  1. Add a clear label, like ​“This text was written with the help of AI”.
  2. Have a person check it properly. Someone who knows the topic fact-checks every claim, and a named person takes responsibility for publishing it, with their contact details on your website.

A quick read or a spell-check isn’t enough. And if AI changes the text after it’s been approved, it needs checking again. Honestly, every sustainability claim deserves that check anyway, because EU rules against greenwashing are getting stricter too. Want help? COSH!’s Green Claims Checker flags risky wording, but a person should still review the result before you publish.

Chatbots, recommendations and virtual try-on

BLOG AI act cosh 1
  • Chatbots: If customers chat with an AI assistant on your webshop, tell them at the start, for example: ​“Hi, I’m an AI assistant.” A line in your terms and conditions isn’t enough, because hardly anyone reads those.
  • Recommendations: ​“You might also like” suggestions don’t need a label.
  • Virtual try-on: It doesn’t count as a chatbot under these rules, but if it generates new images, the tool’s provider must mark them as AI-generated. If a tool sorts people into categories based on their biometric data, for example by guessing their age or gender, you must tell them, and privacy law (GDPR) applies too.
  • Emotion-reading technology: Avoid tools that read emotions from faces, such as some in-store camera systems. They count as high-risk, and using them on employees is banned.

Beyond the AI Act: contracts, your images and your customers

The AI Act isn’t the only thing to think about when you use AI:

  • Model contracts: If you use AI to recreate a real model’s face or voice, you need their permission. Many older contracts don’t cover this.
  • Your own images: Don’t want your collections used to train other companies’ AI? Add a machine-readable opt-out to your website; your web developer can set this up. Large AI developers are required to respect it.
  • Telling customers: If your service is built around AI, such as a styling app, explain this before customers sign up.
  • Your suppliers: Ask the companies behind your AI tools how they mark AI-generated output.
  • The brands you stock: If you run a multi-brand store, ask the brands you buy from whether their product images are AI-generated. The brand that created an AI image is responsible for labelling it, and should make sure the label reaches you. If you simply pass on a brand’s images, you’re generally not responsible under the AI Act, but you should keep any labels intact. The real risk lies elsewhere: if an image makes a product look better than it is, your customers will hold you accountable, and consumer law still applies. So compare product images with the items you receive, and if you edit a brand’s image with your own AI tool, the labelling duty becomes yours.

Sources

Legislation and official EU sources

Legal analysis

Disclaimer: This article is general information, not legal advice. The Commission’s guidelines are not legally binding; only the Court of Justice of the European Union can give the final interpretation of the AI Act. For questions about your specific situation, consult a lawyer.

Özlem Türkmen

Legal intern

Share

Related articles

Discover all articles