What does the AI labeling obligation mean for my ordinary business website?
Article 50 of the EU AI Act has applied since 2 August 2026. What you as the operator of an ordinary business website actually need to do – and what's the AI providers' job.
For an ordinary business website the AI labeling obligation is manageable. Article 50 of the EU AI Act has applied since 2 August 2026. For you as the operator it mainly means this: if you publish realistic-looking AI images of people or events – so-called deepfakes – you must label them visibly. The machine-readable mark that every AI image should carry is set by the AI providers themselves; your job is only not to lose it. For clearly recognisable illustrations, icons or lightly retouched photos you usually need nothing at all. (Context, not legal advice.)
What this is about
Article 50 distinguishes two levels. A machine-readable mark in the metadata that identifies an AI image as artificially generated – technically via methods such as C2PA. And a visible label that visitors see directly. The machine-readable level is the AI providers’ responsibility; the visible level concerns the operator in the case of deepfakes.
What is your job – and what is not
Your job: if you publish a deepfake, that is, a realistic-looking depiction that did not happen that way and could deceive, you label it visibly. Not your job: generating the machine-readable mark yourself – Midjourney, DALL·E, Firefly and others already do that when the image is created. Nothing needed at all for obviously artificial or clearly artistic motifs, or for standard editing such as cropping or exposure.
A common misunderstanding
The obvious thought is: just put a small “AI-generated” label on the image with an image editor and be done. That covers the visible level, but it does not replace the machine-readable mark in the metadata. The good news: you do not have to generate that yourself – it is already in modern AI images. You only have to make sure it is preserved.
The WordPress problem
That is exactly where the catch is: when WordPress generates the various image sizes on upload, this metadata is lost in many cases. The front end then usually serves one of those derived sizes – that is, precisely the file without the mark.
What you should do in practice
Label deepfakes visibly. For all other AI images, make sure the existing machine-readable mark is preserved across all image sizes. And when in doubt, let the editorial team decide whether an image needs a visible label.
Where IMG Performer fits in
IMG Performer detects AI images on upload from their metadata and signatures, writes the machine-readable mark back into every image size generated by WordPress, and adds a visible label on request. The details are in the guides C2PA labeling in WordPress and Labeling AI images for the EU AI Act.
Disclaimer
Legal note: general context, not legal advice. The binding sources are the original text of Article 50 of the EU AI Act (Regulation (EU) 2024/1689, eur-lex.europa.eu/eli/reg/2024/1689/oj) and the guidance of the responsible authorities. As of September 2026.
Frequently asked questions
Are there fines?
Breaches of the transparency obligations in Article 50 can be penalised under Article 99 with up to EUR 15 million or 3% of worldwide annual turnover. For an ordinary business website without deepfakes the risk is low as long as the existing mark is preserved.
Do I have to label old AI images retroactively?
No. Images generated before 2 August 2026 are exempt. Voluntary retroactive labeling is recommended.
What about a clearly recognisable AI illustration in a blog post?
It does not fall under the visible disclosure obligation for deepfakes. The machine-readable mark should still be preserved.