AI-generated images have become so convincing that they can easily pass as real photographs in everyday life. The question of how to recognise them and which labels exist matters to students who read news, share posts or create images themselves. This guide explains the labelling obligation and shows where its limits lie.
Short answer: You cannot reliably recognise AI-generated images from their appearance alone. Since 2 August 2026, the labelling obligation under Article 50 of the EU AI Act has applied, though only in certain cases. For deepfakes, the Chamber of Industry and Commerce (IHK) calls for a clear label, such as a caption. A missing label is no proof that a picture is real. Last updated: November 2026.
Key points at a glance
- The labelling obligation for AI content has applied since 2 August 2026, in certain cases according to the IHK.
- Deepfakes should be clearly labelled, for example with a short caption.
- For content placed on the market before 2 August 2026, the IHK names a transition period until 2 December 2026.
- A missing label does not prove that an image is genuine.
- Classification becomes reliable only through the source, context and origin of an image.
What does the labelling obligation require?
According to the IHK, the obligation covers content that has been created or altered by AI. Its core is that realistic-looking representations must be clearly labelled. The IHK page does not expressly separate the duties of providers and deployers for images, audio and video. It does note that the role may bring additional requirements or different interpretations.
| Case | According to the IHK |
|---|---|
| Deepfake, meaning AI content that looks authentic | Must be clearly labelled |
| Image with a label | A short caption or a note in the image description is enough |
| Video | A note in the description or a short text in the picture, optionally with an EU label |
| Graphic illustrations, caricatures, comics, icons | Named as an exception |
| Advertising illustrations | Named as an exception |
| Photos with only slight changes, such as sharpening or colour correction | Named as an exception because they are not materially altered |
The exceptions matter, but they do not apply automatically. An image does not become harmless merely because it looks graphic if it deceives about a place, a person or an event. The IHK also names cases in which insignificant elements in the background are removed without changing the content.
Why labelling is not the same as recognition
A label helps only if it is present and has not been removed. The IHK description does not mention metadata or watermarks as means of labelling. You also cannot see from an image whether it carries such markers. A label is therefore a hint you can take into account, but not a test result. It also does not replace your own check when a picture makes a claim about real people or events.
Detection tools deliver an assessment as well. A result is not certainty, and it can change as image generators develop. Check several indicators at once before you classify a picture as genuine or artificial.
An everyday example
In a social media post you see a photograph of a demonstration that shows a dramatic event but names no source. Before you share the post, you search for the original posting, check whether reputable media use the picture, and look for labels. If you cannot find an origin, you leave the picture out. A post that works mainly through a picture deserves particular caution.
Why timing and origin matter
An image that first appears on an unknown site deserves more scrutiny than one from a known archive. Note where you found the picture and compare it with the earliest verifiable copy. Images are often taken out of their original context and given a new message. A look at date and place therefore often helps more than asking whether the picture looks artificial.
What you can watch for in everyday life
The following signals are indicators. Each one can also appear with genuine photographs, so they carry weight only in combination.
| Signal | What it can mean |
|---|---|
| No recognisable source | An image without an origin is hard to place and should be treated with caution |
| Inconsistent details in hands, lettering or background | A hint, but no proof, since modern images often look convincing |
| Emotionally charged depiction without evidence | Looking at the source matters especially for news and social media |
| Missing label | Can be a breach, but can also concern an exception |
What applies if you publish AI images yourself?
If you publish AI images yourself, for example in a presentation, on a website or in a social media post, you should check the labelling under Article 50. Whether the obligation applies in an individual case depends on whether the image looks authentic and on how it is used.
Your university rules also apply. Examination regulations and submission rules can set their own requirements for the use of AI. Before you hand in work, check the rules of your programme.
Frequently asked questions
Can I recognise AI images reliably?
There is no reliable recognition from the image alone. The source, the context and the question of who published the image first are helpful. A detection tool gives an assessment, but not proof.
Is a caption enough as a label?
According to the IHK, a short caption or a note in the image description is enough for images. For videos, a note in the description or a short text in the picture is provided for. Whether a label suffices in an individual case depends on the context.
Does the labelling obligation apply to art and illustrations?
For graphic representations such as illustrations, caricatures or comics, the IHK names an exception. Whether a particular work falls under it depends on the case, in particular on whether it deceives about a place, a person or an event.
What applies to images that were online before August 2026?
According to the IHK, content placed on the market before 2 August 2026 has a transition period until 2 December 2026. Check the position in your specific case before drawing conclusions.
How should I handle a questionable image?
Check the origin, look for the original post and do not share the image until you have classified it. In political or personal contexts, restraint makes sense, because false images can cause harm quickly.
How to proceed now
- For every striking image, check the source and the context first.
- Do not rely on a label or a detection tool alone.
- Label your own AI images if the obligation applies in your case.
- Clarify your university rules on AI use before you submit work.
Related reading
- AI bans in the EU: Which applications have been prohibited since February 2025 (in German)
- Digital rights in everyday life: An overview for students
- Check app permissions (in German)
- Consumer rights in everyday life
Last updated: November 2026. This guide is general information, not individual advice.
