EU AI Act from 2 August 2026: what the AI-labelling rule means for online sellers

On 2 August 2026, a part of the EU AI Act goes live that concerns anyone using AI images or AI videos for their shop today: the transparency obligations in Article 50. From that date, artificially created or altered media must in principle be labelled so that their AI origin is detectable by machines.

That sounds like a topic for corporate legal departments. It is not. AI-generated product images and videos have long been everyday practice in e-commerce — from the cut-out packshot to the model wearing your garment. These are exactly the assets the new rule covers. This article explains — calmly and without scaremongering — what changes, who is affected, and why the solution is more technical and more discreet than the headlines suggest.

What 2 August 2026 actually means

The AI Act entered into force on 1 August 2024. Its obligations apply in stages. The transparency rules for AI-generated content under Article 50 belong to the stage that becomes applicable 24 months later — on 2 August 2026.

The core in one sentence: anyone providing an AI system that generates image, video or audio content must ensure those outputs are marked in a machine-readable format as artificially generated or manipulated. It is explicitly about a marking that machines detect — not necessarily something a person sees at first glance.

Which is the most important reassurance up front: no visible watermark on your product image is required.

Why this concerns online sellers in particular

Two years ago, "AI image" was a niche topic. Today, sellers use it for routine assets: swapping backgrounds, cutting out products, dressing a garment on an AI model, turning a photo into a short product video. These assets land on product pages, in marketplace listings and on social channels — exactly where consumers make buying decisions.

The AI Act distinguishes two roles here:

• Providers of the AI systems that generate the content. They carry the technical duty to mark the outputs in a machine-readable way.
• Deployers who use certain content — in particular convincingly realistic "deepfakes" of people, or content on matters of public interest — with their own disclosure duties.

For the classic case of an "AI-generated product image", the technical lever therefore sits with the system that creates the image. For you as a seller that means: what matters is that the tool you make your images and videos with labels correctly. Whether any further disclosure duties apply to you in a given case depends on the specific content and, in case of doubt, is a question for qualified legal advice — this article does not replace it.

Not a watermark — a provenance record inside the file

The obvious but wrong assumption is: "So now a visible AI label has to go on the image." For a sales image that would be absurd — a stamp over the product costs conversion and helps no one.

The established route is different, and it is called C2PA (Coalition for Content Provenance and Authenticity), in consumer terms Content Credentials. Behind it stands an open industry standard, backed among others by Adobe, Microsoft, the BBC, Sony and many more. The idea:

1. When the content is created, a cryptographically signed provenance record is written directly into the image or video file.
2. This record states, machine-readably: This content was created with AI. — including the note that it is "trainedAlgorithmicMedia".
3. The visible image changes nothing. No stamp, no bar, no quality loss.
4. Verification tools — such as the public contentcredentials.org — read the signature and display it. Platforms can evaluate it automatically too.

This kind of marking is exactly what Article 50 means by "machine-readable format". It fulfils the transparency idea without disfiguring your marketing asset.

What you should do now

Three sober steps, without alarmism:

1. Get an overview of where you use AI. Product images, cut-outs, model shots, product videos — labelling is relevant everywhere there.
2. Check whether your tool labels correctly. The decisive question to any image or video provider is simply: Do you write C2PA provenance information into the output files? An easy test: upload a generated image to contentcredentials.org and see whether a signature appears.
3. Clarify your own disclosure duties with an expert if in doubt. Especially if you work with realistic depictions of people, a disclosure beyond the technical marking may be advisable or required.

How Buust solves this

Since July 2026, Buust automatically labels every AI-generated video and image to the C2PA standard. The provenance information travels invisibly in the file information — no watermark on the image, no quality loss, no extra click and no surcharge. You create your videos and images as before; the labelling happens in the background.

This is built deliberately that way: the label belongs in the metadata, not as a stamp on the sales image. That settles the technical part of the transparency requirement for all content created in Buust — and it does so before 2 August, not at the last minute.

If you want to see how this looks in practice: any product video created in Buust can be dropped into contentcredentials.org, where it shows the provenance record "AI-generated product video". The label thus becomes visible exactly where it belongs — in the verification tool, not on your customer's image.

---

This article is general guidance and not legal advice. For a binding assessment of your specific case, consult qualified legal counsel.