The time has come: As of August 2, 2026, new transparency requirements for AI-generated content will take effect, requiring companies to label such content. Contrary to many oversimplified accounts, however, the labeling requirement is much more limited than one might think. We’ll explain the requirements and exceptions to you.
AI Labeling Becomes Mandatory
AI now creates text for websites, generates product images, edits photos, and produces videos for social media—or assists with these tasks, as is the case with this newsletter. However, these new capabilities also come with disclosure requirements.
As of August 2, 2026, the transparency requirements of Article 50 of the AI Regulation apply to certain content. They are intended to prevent people from mistaking AI-generated or manipulated content for genuine content.
Particularly relevant: images, videos, and audio files
In practice, this labeling is likely to be particularly relevant for images and videos. For example, if a photo of a person, object, location, facility, or event that closely resembles reality is generated by AI or altered in a way that deviates from reality for an advertising campaign, it must be labeled as AI-generated or manipulated content. The same applies to audio files.
With such “deepfakes,” the information must be presented to the viewer in a clear and understandable way. A hidden note in the metadata—as AI systems already do on their own—is not enough.
In marketing, in particular, it is therefore important to consider how AI-generated images should be labeled in the future. To this end, the European Commission has developed voluntary symbols for labeling AI-generated content, which are either general in nature or depict the specific type of processing involved:
When it comes to texts, it all depends on when they are published
The situation is more nuanced when it comes to AI-generated texts. The AI Regulation establishes a transparency requirement for AI-generated or manipulated texts that are published to inform the public about matters of public interest.
Therefore, a company is not automatically required to label every text on current events that was generated using AI with an AI disclosure. If AI is used to create an extensive article on a socially or politically relevant topic in its entirety, and the article is subsequently published essentially unchanged, there is a disclosure requirement. Conversely, there is no obligation to label the text if it is either not published as an editorial medium or has been subject to appropriate editorial review and a natural or legal person assumes editorial responsibility. Therefore, the focus of the regulations is not on the AI-assisted creation of any texts, but rather on automated, uncontrolled, and thus potentially inaccurate media reporting.
Relevance to Daily Work
Companies should therefore not only regulate the use of AI, but also specify what content is created with it and to what extent. For example, a company that uses AI solely for spell-checking, translation, or refining individual phrases is in a different situation than a company that uses it to generate entire advertising images, videos, or editorial content—or even publishes them automatically.
The following aspects should be helpful in this regard when conducting an internal assessment:
- Which AI tools are permitted—for example, from a data protection perspective? Is there a guideline on this?
- Develop a clear framework for usage and ensure your employees adhere to it.
- What kind of content is created using this? Has it been reviewed or edited by a human?
- For texts, human editing or review is often sufficient to avoid being subject to labeling requirements.
- In the case of images, videos, and audio recordings, the key factor is whether they depict people, objects, places, or events that resemble reality but did not actually take place.
Conclusion
The new transparency requirements do not mandate blanket labeling of all AI uses. However, when it comes to AI-generated images, videos, and audio, as well as automatically published text, companies should adapt their existing processes.
