AI changes images – but does it need to be labeled?

Jennifer Schülzky
by Jennifer Schülzky · 27.09.2026
Hand ticking a box in an online form on a laptop

Two laws – two different questions

Whether an image has been manipulated with artificial intelligence is not, at first, a data protection issue.

The GDPR deals with this,

  • whether personal data is processed,
  • whether there is a legal basis for this
  • and what rights the affected individuals have.

The AI Act, by contrast, pursues a different goal.

This is about creating transparency in the use of artificial intelligence. Under certain conditions, viewers should be able to recognize when content has been generated or significantly modified using an AI system.

Therefore, it is not enough to only look at the GDPR or the AI Act.

When does the transparency obligation under Art. 50 AI Act apply?

Art. 50 AI Act distinguishes between the transparency obligations of providers and those of deployers of AI systems. Providers of certain generative AI systems must ensure that artificially generated or manipulated content is, in principle, marked in a machine-readable format and recognizable as such.

For companies that deploy such systems, Art. 50(4) AI Act is particularly relevant. Under this provision, deployers must disclose when they use an AI system to generate or manipulate image, audio or video content that qualifies as a so-called deepfake.

A deepfake exists when artificially generated or manipulated content noticeably resembles real persons, objects, places, entities or events and could falsely be perceived as genuine or truthful.

Not every AI-assisted image edit therefore automatically leads to a disclosure obligation. What matters are the specific change, the effect on viewers and whether the content qualifies as a deepfake within the meaning of the AI Regulation.

What exactly is a significant change?

It is here that most misunderstandings are currently arising.

Not every image manipulation is automatically an AI manipulation within the meaning of the AI Act.

Typical examples that occur regularly None significant changes that may occur include:

  • Adjust brightness,
  • Change the contrasts,
  • Optimizing colors,
  • Remove image noise,
  • minor retouching
  • or cropping a picture.

It is different, for example, when

  • People removed or added,
  • Faces exchanged,
  • Background completely replaced,
  • Items inserted
  • or realistic image content is recreated.

The deciding factor for the deployer’s disclosure obligation is not only how extensively an image has been changed with artificial intelligence. Rather, what matters is whether the editing creates content that noticeably resembles real persons, objects, places or events and could falsely be perceived as authentic.

Even minor changes can therefore be legally relevant if they create a misleading impression. Conversely, not every extensive AI edit automatically leads to a disclosure obligation under Art. 50(4) AI Act.

Our practical example from the previous post

An employee revokes their consent to publish a group photo. The company decides not to completely delete the image but to remove the person from the photo using an AI image editor.

From a GDPR perspective, the first step is to check whether the data subject is still identifiable in the edited photo. If the person is removed completely and identification is no longer possible even on the basis of other available information, the edited image may no longer contain personal data in this respect. However, the processing of the original photo and any personal data that remain stored must be assessed separately.

However, this does not necessarily conclude the review. Regardless of this, a second question arises: does the AI editing lead to a transparency obligation under Art. 50 AI Act?

If, for example, an employee is removed from a group photo using artificial intelligence, it must be checked whether the edited image qualifies as a deepfake within the meaning of the AI Act. A labeling obligation does not exist merely because a person has been removed. However, if the edited photo creates a false impression of a situation that actually took place, a disclosure obligation under Art. 50(4) AI Act may apply.

The removal of a single person does not automatically result in a significant change in the meaning of a picture. Otherwise, this may appear when the editing alters the image in a substantive way or conveys a different overall impression. This could be the case, for example, when people are added or replaced, entire scenes are altered, or completely new image content is created. In these cases, it should be carefully examined whether the transparency requirement under Art. 50 AI Act applies.

When must it be marked?

Companies should, in particular, consider,

  • whether an AI system was used to generate or manipulate the image,
  • whether the result noticeably resembles real persons, objects, places or events,
  • whether viewers could falsely consider the generated or manipulated content to be authentic,
  • whether this constitutes a deepfake within the meaning of Art. 3(60) AI Act,
  • and whether its disclosure obligation or a statutory exception under Art. 50(4) AI Act applies.

As a general rule, the stronger the AI’s ability to alter the content of an image and the more realistic the result appears, the more likely it is to be examined whether there is a transparency obligation under Art. 50 of the AI Act. Pure technical optimizations such as brightness or color corrections will often be evaluated differently than the removal or replacement of people or the creation of entirely new image content.

GDPR and AI Act Compared

Does an AI-generated image need to be labeled?

Note: This infographic was created with the support of artificial intelligence and subsequently reviewed and revised by our data protection and AI experts.

The graphic is intended as a first guide and does not replace a legal assessment of each individual case. Especially when it comes to marketing campaigns or publicity-oriented publications, a careful evaluation of the specific image manipulation is recommended.

What companies should now consider

From our consulting practice, it is particularly advisable to,

  • Data protection and the AI Act should be evaluated separately,
  • documenting the use of AI for image processing,
  • to check whether individuals are still identifiable after processing,
  • To raise awareness among marketing and HR departments regarding the new transparency requirements,
  • introduce binding internal processes for the release of AI-processed images,
  • and define responsibilities for the labeling of artificially modified content.

Since AI image processing functions are now integrated into many standard programs, appropriate changes are often made without sufficiently considering the legal implications.

The transparency obligations under Art. 50 AI Act have applied since August 2, 2026. Companies should therefore review existing processes for AI-assisted image editing and take necessary disclosures into account when publishing such content.

Conclusion

The integration of AI image processing into common software facilitates numerous workflows for companies. At the same time, it creates new data protection and regulatory issues that are often underestimated.

In short: The GDPR answers the question of whether personal data may be processed. The AI Act answers the question of whether the use of artificial intelligence against third parties must be made transparent.

Our experience shows that these two examinations are often mixed together. Those who cleanly separate the two regulatory areas from each other and incorporate the relevant requirements early on into existing processes create more legal certainty and reduce the risk of later adjustments.

Check the mandatory fields –
Step-by-step in the webinar

FAQ

Do AI-edited images always need to be labeled?


What is the difference between the GDPR and the AI Act regarding AI-processed images?


Which image modifications are not considered to be a significant change on a regular basis?

When can AI image processing be essential?


Does a group photo have to be labeled if a person was removed using AI?


What should companies take into account organizationally when using AI image processing?