AI content labelling obligation from August 2026: what creators and businesses need to know

From 2 August 2026, businesses and individuals in the EU must clearly label certain AI-generated content, or face fines of up to €15 million or 3% of global annual turnover (Europäische Union, 2024; Weventure, 2026). Anyone already working with ChatGPT, Midjourney, or AI image tools and publishing that content should take a close look now, not wait until 1 August.

The labelling obligation is part of the EU AI Regulation (the AI Act), which entered into force on 1 August 2024 but applies in stages (Europäische Union, 2024). The specific transparency obligations are set out in Article 50 and become binding from 2 August 2026 (EU Artificial Intelligence Act, 2026a; McCann FitzGerald, 2026). The aim of the rule is that people can recognise when they are interacting with AI or consuming AI-generated content, protecting them from deception and manipulation (Europäische Union, 2024).

A note on timing, because this is moving fast: at the time of writing, the European Commission's detailed guidelines on Article 50 were not yet final, nor was the accompanying Code of Practice on labelling AI content (Europäische Kommission, 2026a, 2026b). A delay to certain technical obligations, agreed between the EU Council and Parliament in May 2026, was at the time of writing still only a provisional political agreement, not yet formally adopted in the EU Official Journal (Sidley Austin LLP, 2026). Details may therefore still shift before the deadline. For current information, the European Commission's official page on the Code of Practice is worth checking (Europäische Kommission, 2026a).

AI content labelling obligation: What Article 50 actually requires

The AI Regulation distinguishes four situations in which transparency becomes mandatory (EU Artificial Intelligence Act, 2026b; Sidley Austin LLP, 2026).

Interacting with AI systems.Anyone offering a chatbot, voice assistant, or AI companion must ensure that users can recognise they are talking to a machine. This does not apply where the AI nature is obvious to a reasonably attentive person anyway.

Synthetic content.Providers of AI systems that generate audio, image, video, or text must ensure their outputs are marked in a machine-readable format, for example through watermarks or metadata, so that detection tools can identify them as AI-generated (EU Artificial Intelligence Act, 2026a).

Emotion recognition and biometric categorisation.Anyone using such systems must inform the people affected.

Deepfakes and text on matters of public interest.Anyone publishing image, audio, or video content that appears deceptively real (classic deepfakes) must clearly and visibly label it as such. For text on matters of public interest, such as news or political content, the obligation only applies where no person takes editorial responsibility for it (Haufe, 2026; WKO, 2026).

This last point is usually the most relevant one in practice: an AI-generated text that a responsible person reads, edits, and approves does not, according to several consistent legal analyses, need to be labelled. If it is published automatically and unchanged, it does (Haufe, 2026; Wirtschaftskammer Österreich, 2026). Advertising copy, product descriptions, or social media posts unrelated to matters of public interest generally do not fall under the text-labelling obligation as things currently stand, whereas pure image, video, and audio deepfakes almost always do (Wirtschaftskammer Österreich, 2026).

Photorealistic images and video: where the most action is needed

For anyone working extensively with AI image and video content, this is exactly where the regulation bites hardest. Article 50 distinguishes two layers of obligation, and the two are frequently confused in practical guides (Legal Tribune Online, 2026; meedia.de, 2026). Layer one is the provider's obligation: tools such as Midjourney, Adobe Firefly, or Google Gemini must mark their outputs in a machine-readable way, for example through invisible watermarks such as Google's SynthID or metadata standards such as C2PA. That is the tool makers' job, not yours (meedia.de, 2026).

Layer two concerns you directly: anyone who publishes an AI image or video professionally counts legally as a deployer and must additionally attach a label visible to humans wherever the content qualifies as a deepfake. An invisible watermark alone is explicitly not sufficient, even if it is technically embedded correctly (Legal Tribune Online, 2026; Digitalzentrum Chemnitz, 2026; meedia.de, 2026). The regulation's definition of a deepfake is also broader than many assume: it covers not just malicious manipulations of real people, but any AI-generated or AI-altered image, audio, or video content that resembles real people, places, or events and could be mistaken for genuine (Legal Tribune Online, 2026). A photorealistic AI portrait of a fictional person in an advert falls under this just as much as an edited photo of a real location.

According to a consistent view across several sources, stylised, clearly recognisable artistic illustrations with no connection to real people or places are not affected, nor is ordinary AI-assisted retouching or colour correction, as long as it does not substantially change the meaning of the image (Marketing-KI.de, 2026; webdesignen.at, 2026). Exactly where the line between minor correction and substantial change falls, for example with generative fill or sky replacement, was not yet conclusively settled as of the Commission's draft guidelines from May 2026 (Marketing-KI.de, 2026). When in doubt: label rather than risk it.

The label itself must sit on or immediately next to the image, for example as a caption or a small tag directly on the image, not in the alt text, the footer, or the terms and conditions (Digitalzentrum Chemnitz, 2026; webdesignen.at, 2026). In June 2026, the European Commission published standardised, free icons for this purpose ("AI", "AI GENERATED", "AI MODIFIED") that can be used directly, though this may still change before the final version of the Code of Practice (webdesignen.at, 2026).

For everyday practice, a simple immediate step is worth adopting, regardless of the tool used: create a dedicated folder for AI images. Mark every file name with the suffix "_AI", "_KI", or similar. Save the prompt used as a text file alongside it. This keeps every image traceable, which is also what legal advisers recommend to agencies (Marketing-KI.de, 2026). For photorealistic AI images, additionally add a human-visible label at the point of publication in line with Article 50 of the AI Regulation. This documentation costs a few seconds per image but saves a lengthy investigation if it is ever needed.

Does this apply in Switzerland too?

Formally, no. In practice, mostly yes. Switzerland is not an EU member state, and the AI Regulation is not Swiss law. There is currently no Swiss legislation of its own requiring AI content labelling (Bundesamt für Justiz, 2025). In February 2025, the Federal Council decided against a horizontal AI law modelled on the EU's, opting instead for targeted adjustments to existing legislation and ratification of the Council of Europe's AI Convention. A corresponding consultation draft is expected by the end of 2026 (Bundesamt für Kommunikation, 2025; Bundesamt für Justiz, 2025).

Even so, the EU rule reaches many Swiss businesses through the market-location principle: anyone producing content for EU customers or addressing EU users falls under the AI Act for that part of their activity, regardless of where the company is based (Knowlee, 2026). For Swiss businesses operating purely domestically, labelling is not yet a legal requirement, but given the planned Swiss regulation and the international orientation of many Swiss SMEs, it is a sensible step to take early.

Fines and platform rules

Violations of Article 50 can be penalised with fines of up to €15 million or 3% of global annual turnover, whichever is higher (Europäische Union, 2024; Weventure, 2026). By comparison, the highest fine tier under the AI Regulation, up to €35 million or 7%, applies to prohibited AI practices and serious breaches of the obligations for the most capable AI models, not to the transparency obligations under Article 50 (Regulations.ai, 2026).

Regardless of the legal position, some platforms already label AI content themselves. LinkedIn, for instance, has marked detected AI images using the C2PA standard since 2024, similar to Meta's "Made with AI" labels on Instagram, Facebook, and Threads (MediaPost, 2024). As platform policies change faster than laws, it is worth checking the current community guidelines of the relevant platform before your next campaign.

How to prepare

1. Create a dedicated folder for AI images, mark every file with the suffix "_AI", "_KI", or similar, and save the prompt as a text file alongside it. This single step alone turns the later visible labelling into a formality rather than an investigation.

2. Appoint one person per content process to read AI-generated text before publication and give editorial sign-off. This is often what determines whether a labelling obligation applies at all.

3. Define a standard labelling wording, for example "This content was created with the support of AI", and embed it in the blog footer, social media templates, and chatbot greetings. For images, the label belongs directly on the image or in the caption, not in the alt text.

4. Draw up a list of all AI image, video, and audio tools used in the business, and check which of them can produce deepfake-capable content.

5. Label chatbots and AI assistants on the website unmistakably as such, at the first point of contact, not in the small print.

6. Keep an eye on the European Commission's final guidelines and Code of Practice through to August, as details may still be refined (Europäische Kommission, 2026a).

My personal assessment

In my view, this labelling obligation has a structural weakness: it mainly reaches those who already work carefully and transparently. Actors with criminal intent who use deepfakes for fraud, disinformation, or fake testimonials are unlikely to pay much attention to a labelling requirement. In practice, the rule therefore tends to affect honest users more than the actors it is actually meant to guard against, adding administrative effort for them in the process.

Even so, I consider the underlying approach sound and the rationale understandable. For the vast majority of professional AI use, a consistent labelling standard creates more clarity in the digital space and makes visible where humans, and where machines, have been at work.

This does not solve the underlying problem of misuse, but it is a meaningful contribution to greater transparency. A clearly labelled piece of AI content is not a flaw, it is a "proof of human" at exactly the point where I have taken editorial responsibility, and an honest disclosure where I have not. To me, Human First does not mean hiding AI, it means openly showing where the human stands in the process.

Concrete recommendations for the coming weeks

1. Set up your AI image folder with a file-name suffix this week and document the corresponding prompts, this is the fastest and most effective single step.

2. Do a quick stocktake by the end of July: which content is genuinely AI-generated, and which is only AI-assisted?

3. Define editorial responsibility in writing, this is your most important lever for exemptions from the text-labelling obligation.

4. If you have an international audience, check whether EU customers are affected, even as a Swiss business.

5. Use labelling actively as a trust signal in your communication, rather than hiding it.

References

Bundesamt für Justiz. (2025). Artificial intelligence. https://www.bj.admin.ch/de/kuenstliche-intelligenz

Bundesamt für Kommunikation. (2025). Artificial intelligence. https://www.bakom.admin.ch/de/kuenstliche-intelligenz

Digitalzentrum Chemnitz. (2026, 20 January). Labelling AI-generated images and text [KI-generierte Bilder und Texte kennzeichnen]. https://digitalzentrum-chemnitz.de/wissen/ki-generierte-bilder-und-texte-kennzeichnen/

EU Artificial Intelligence Act. (2026a, 14 May). The EU AI Act's transparency rules: A practical guide to Article 50. https://artificialintelligenceact.eu/transparency-rules-article-50/

EU Artificial Intelligence Act. (2026b). Article 50: Transparency obligations for providers and deployers of certain AI systems. https://artificialintelligenceact.eu/article/50/

Europäische Kommission. (2026a). Code of practice on transparency of AI-generated content. Digital Strategy. https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content

Europäische Kommission. (2026b, 8 May). Consultation on the draft guidelines on transparency obligations under the AI Act. Shaping Europe's Digital Future. https://digital-strategy.ec.europa.eu/en/consultations/consultation-draft-guidelines-transparency-obligations-under-ai-act

Europäische Union. (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (AI Act). Official Journal of the European Union, L, 2024/1689. https://eur-lex.europa.eu/eli/reg/2024/1689/oj

Haufe. (2026). Labelling obligation for AI content applies from August 2026 [Kennzeichnungspflicht für KI-Inhalte gilt ab August 2026]. https://www.haufe.de/recht/kanzleimanagement/kennzeichnungspflicht-fuer-ki-inhalte-gilt-ab-august-2026_222_681220.html

Knowlee. (2026, April). The EU AI Act in Switzerland: Who it applies to, and who it does not [EU AI Act in der Schweiz]. https://www.knowlee.ai/ch/blog/eu-ai-act-schweiz-anwendbarkeit

Legal Tribune Online. (2026, 1 July). Labelling obligation for AI content: what applies to deepfakes and more? [Kennzeichnungspflicht für KI-Inhalte] (K. Schreiber, Author). https://www.lto.de/recht/kanzleien-unternehmen/k/transparenz-kennzeichungspflicht-fuer-ki-inhalte

Marketing-KI.de. (2026, 6 July). Labelling obligation for AI images: what applies from 2 August 2026 [Kennzeichnungspflicht KI-Bilder]. https://marketing-ki.de/aktuelles/ki-bilder-kennzeichnungspflicht-ab-2026/

McCann FitzGerald. (2026). One month to go: EU AI Act transparency compliance. https://www.mccannfitzgerald.com/knowledge/technology-and-innovation/one-month-to-go-eu-ai-act-transparency-compliance

meedia.de. (2026, 5 June). AI labelling from August: what businesses need to decide now [KI-Kennzeichnung ab August]. https://meedia.de/news/beitrag/24238-ki-kennzeichnung-ab-august-was-unternehmen-jetzt-entscheiden-muessen.html

MediaPost. (2024, 22 May). LinkedIn begins labeling AI-generated content. https://www.mediapost.com/publications/article/396225/None

Regulations.ai. (2026, 6 January). Regulation (EU) 2024/1689, Artificial Intelligence Act (EU AI Act). https://regulations.ai/regulations/RAI-IE-NA-E2AIEXX-2024

Sidley Austin LLP. (2026, 24 June). EU AI Act transparency obligations: Preparing for compliance by 2 August 2026. Data Matters Privacy Blog. https://datamatters.sidley.com/2026/06/24/eu-ai-act-transparency-obligations-preparing-for-compliance-by-2-august-2026/

webdesignen.at. (2026). AI content labelling obligation under Article 50 EU AI Act from August 2026 [KI-Kennzeichnungspflicht nach Artikel 50]. https://www.webdesignen.at/blog/ki-kennzeichnungspflicht-2026-webseiten/

Weventure. (2026, 7 May). Labelling obligation from 2026 for AI-generated content [Kennzeichnungspflicht ab 2026]. https://weventure.de/en/blog/ai-labeling

Wirtschaftskammer Österreich. (2026). Labelling obligation for certain AI content from August 2026 [Kennzeichnungspflicht für bestimmte KI-Inhalte]. https://www.wko.at/gewerbe-handwerk/kennzeichnungspflicht-fuer-ki-inhalte

---

About the author

Harriet Moser is an AI expert specialising in branding and the founder of Ask Harriet, a human-led AI training consultancy based in Switzerland. As part of her master's thesis, she developed the AI Brand Strategy Framework for integrating artificial intelligence across five levels of brand management. She specialises in helping businesses and individuals use AI systematically, ethically, and effectively.

For more information: www.askharriet.ch | LinkedIn: linkedin.com/in/harriet-moser

---

DISCLAIMER: I am not a lawyer. This article is for educational and informational purposes only. While every effort has been made to ensure accuracy at the time of publication (July 2026), the legal situation around the AI Regulation is currently evolving quickly, particularly the not-yet-finalised guidelines and Code of Practice for Article 50. This post should therefore not be construed as legal advice. For a binding assessment of your situation, please consult a lawyer specialising in technology law.