The EU Artificial Intelligence (AI) Act requires companies to disclose in certain cases whether content has been created or modified by AI. This also applies to companies that make use of AI such as image-generation programs for product advertising.
Given that the new rules are enforceable from 2 August 2026, with fines up to EUR 15 million or 3% of global annual turnover, the importance of compliance with these rules is now in sharp focus.
The good news is that for deployers of AI systems, besides chatbot disclosure, labelling obligations only apply in two scenarios: (i) AI-generated or manipulated texts aiming at informing the public on matters of public interest, and (ii) deep fakes[1].
Looking at all of this in more detail, what does this mean to companies that create and/or publish advertising and promotional content?
Why AI-Generated or Manipulated Texts on Public Matters Might be a Concern
As one example of AI generated or manipulated texts which inform the public on matters of public interest, the EU Commission’s Draft Guidelines (the latest version being a 2024 draft for stakeholder consultation)[2] list AI-manipulated corporate reports published on a listed company’s website which contain investor information.
By contrast, the EU Commission’s Draft Guidelines assume that the labelling obligation will not apply to:
- AI-manipulated texts that are part of a company’s advertisements other than text that promotes any claims relating to e.g. health, consumer safety or sustainability
- AI-generated fictional novels or poems in any genre
- news summaries by a chatbot that are only available to the user that prompted the chatbot.
A general exemption applies to AI generated or manipulated text that has undergone human review or editorial control and for which a legal or natural person holds editorial responsibility[3].
Deep Fakes
What seems more challenging for companies promoting or selling products on their websites, in catalogues, on billboards, on TV and social media or other means of advertising, is the obligation to disclose AI generated or manipulated image, audio or video content constituting a deep fake. The term “deep fake” is defined in Article 3(60) of the AI Act as
“AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful”.
Whilst there are good reasons for extending the definition of “deep fake” beyond persons, i.e. realistic natural human beings, in view of the dangers involved with simulated events without persons being involved, the inclusion of:
- objects,
- places,
- entities, and
- events
opens the door for the labelling obligations applying to a broad spectrum of content including product advertisements.
The key question in the individual case is whether such content is likely to deceive or mislead a person regarding its authenticity or truthfulness. This assessment is to be conducted based on a hypothetical average person expected to be exposed to such content, which is particularly important if advertising is directed at children or groups with lower digital and AI literacy[4].
The sheer variety of options to generate and manipulate content with AI leads to various borderline cases. For example, does color correction, re-scaling, background editing, removal of elements, simulated movement, lighting or seasonal adjustment in relation to product images amount to a deep fake? Most likely not if the lack of authenticity is obvious (e.g. in case of blurred backgrounds) or the manipulation has only a minor impact on the consumer’s perception.
However, images of persons who are made to resemble a celebrity might be caught. The same question may arise in relation to depictions of real persons, for example, an advertisement may use a genuine photograph of a person while AI is merely used to place that person in a kitchen setting where the original image is not taken in a kitchen, or AI is used to change the person’s posture or the shade of the hair.
Practical Considerations
AI-generated or AI-modified content must be clearly labelled in a manner that is easily noticeable to users and remains visible when the content is viewed or shared. Until there is case law available on how strict the labelling requirements are interpreted and how they are fulfilled, the EU AI Office’s Code of Practice on Transparency of AI-Generated Content[5] provides valuable, albeit non-binding practical guidance on the topic.
- Standardized Icons. This includes freely available standardized EU icons that companies may want to use to comply with the labelling requirements.
- Size of Icons. As far as the icon’s size is concerned, the only guideline so far is basically that it must be “clear and distinguishable”[6].
- “AI”, “AI Generated” or “AI Modified”. Whether the designation “AI” is sufficient or – as “encouraged” in the Code of Practice (i.e. not binding but recommended) – transparency requires distinguishing between “AI Generated” and “AI Modified”, is still an open question.
- Implementation of a second layer. According to the Code of Practice, the integration of interactive second layers to enrich the static icons with provenance data and information on what has been modified is currently explored.
- Placing icons on images and videos. For images and videos, the disclosure should be displayed prominently within the visible content area and embedded in the metadata; where available, AI watermarks or visual labels should also be used.
- Audio labelling. Audio content must include a clear audible disclosure.
- Periodic disclosure. Video and audiovisual content must be labelled at the beginning and, for content exceeding a short period of time, continuously or at regular intervals, with both visual and audio disclosures where applicable.
In view of the legal consequences of violations and sometimes huge amount of content involved, it is no wonder that some prominent retailers have already decided to take the bull by the horns and label any AI created or modified images. All of this will be of interest to manufacturers, retailers and the creators of advertising material. Besides the risk of fines, the new labelling requirements will most likely be deemed provisions intended to regulate market behavior under the laws of unfair competition, opening the door for claims by competitors and watchdog associations, with the consequence that advertising may, by way of a preliminary injunction, be prohibited literally overnight. Companies are therefore well-advised to establish internal guidelines and procedures for AI transparency and closely monitor the developments in the sector to verify on a regular basis whether existing compliance measures require updates.
[1] Article 50(4) AI Act.
[2] The consultation period ended on 3 June 2026.
[3] Article 50(4), subparagraph 2 AI Act.
[4] Cf. Draft Guidelines, p. 28.
[5] https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content.
[6] Article 50(5) AI Act.
Disclaimer: While every effort has been made to ensure that the information contained in this article is accurate, neither its authors nor Squire Patton Boggs accepts responsibility for any errors or omissions. The content of this article is for general information only, and is not intended to constitute or be relied upon as legal advice.