Labelling AI Content: What the New EU Icons and the AI Act Require from August 2026

In June 2026, the European Commission published official icons for labelling AI-generated content. At the same time, an important date is approaching: on 2 August 2026, the transparency obligations of the EU AI Act (Article 50) become applicable. Time for a sober assessment — because plenty of half-truths are circulating around this topic.

What actually applies from 2 August 2026

The EU AI Act has been in force since August 2024, but its obligations take effect in stages. On 2 August 2026, Article 50 becomes applicable — the transparency obligations for AI systems that interact with people or generate content. Four scenarios are regulated:

1. Chatbots and AI assistants: Anyone interacting with an AI system must be able to recognise it — unless it is obvious.
2. Machine-readable marking: Providers of AI systems that generate synthetic images, videos, audio or text must mark the outputs as AI-generated in a machine-readable way (for instance via metadata or watermarks). This obligation falls on the makers of the AI tools — such as Google, OpenAI or Anthropic — not on their users.
3. Deep fakes: Anyone publishing AI-generated or AI-manipulated image, audio or video content that deceptively resembles real persons, places or events must disclose that the content was artificially created or altered.
4. AI texts on matters of public interest: Texts that inform the public on matters of public interest and are fully AI-generated — without human review and without editorial responsibility — must be labelled as AI-generated.

For website operators, points 3 and 4 are the most relevant. They are so-called deployer obligations: they fall not on the maker of the AI tool, but on whoever uses and publishes the content.

The three new EU icons

The Commission has presented three standardised icons intended to harmonise labelling:

• Base icon "AI" — generic indication of AI involvement
• "AI Generated" — the content was fully created by AI (apart from the user's prompt)
• "AI Modified" — existing content was altered by AI

The icons are available in four variants (black, white, each with and without transparency) as SVG and PNG for download from the European Commission.

Important: the icons are voluntary. What is mandatory is the labelling itself — the law does not prescribe its form. The icons are part of a Code of Practice which the Commission has assessed as an adequate instrument for implementing the obligations. Anyone using the icons is applying a standard recognised by the authorities.

What must be labelled — and what doesn't

Precision pays off here, because the most common misconception is: "Every AI image now needs a label." That is not true.

Subject to labelling:

• An AI-generated photo showing a real person or looking deceptively authentic (deep fake)
• An AI-generated "customer photo" or testimonial video featuring a non-existent but real-looking person
• AI-cloned voices in audio content
• A fully AI-written news article published without review

Not subject to labelling:

• Illustrations, icons, abstract graphics and stylised hero images — even if AI-generated
• Blog articles created with AI assistance but reviewed and editorially accounted for by a person
• AI-assisted image editing that does not deceptively alter the content (colour correction, background removal, upscaling)
• Artistic, satirical and obviously fictional works (a discreet notice that does not impair the work is sufficient here)

The basic rule: it is about the risk of deception, not about AI use as such. The AI Act aims to prevent people from mistaking artificial content for real — it does not seek to stigmatise every use of AI.

Does this also apply to Swiss companies?

The AI Act is EU law — but like the GDPR, it reaches beyond EU borders. It covers anyone placing AI systems on the EU market or whose AI-generated content is directed at people in the EU. For Swiss companies this means:

• Website (also) addresses EU customers — for instance through deliveries to Germany, EU clientele or targeted marketing in the EU: the transparency obligations apply.
• Purely Swiss audience: the AI Act formally does not apply. Switzerland does not yet have its own AI law; the Federal Council is currently pursuing a sectoral approach guided by the Council of Europe's AI Convention.

In practice, the separation is rarely clean: anyone operating a publicly accessible website in German, French or Italian effectively reaches an EU audience too. Our recommendation follows the proven pattern from data protection: align with the EU standard — then you are on the safe side in both jurisdictions and prepared for future Swiss regulation.

Checklist for SMEs and agencies

1. Take inventory: Where do you use generative AI? Website images, social media posts, product videos, blog texts, newsletters, chatbots?
2. Deep fake check: Is there content featuring real-looking persons, voices, places or events that is AI-generated or AI-altered? This must be labelled from 2 August 2026.
3. Text check: Do you publish fully AI-generated texts without human review — such as automated news feeds or AI product texts at scale? Then label them or introduce an editorial review step.
4. Chatbot check: Is it clear to users that they are talking to an AI? A brief notice in the chat window is sufficient.
5. Define your labelling format: Download the EU icons and add them to your design templates — or choose your own clearly understandable wording ("Created with AI").
6. Agencies additionally: Update client contracts and approval processes — who provides the labelling, who is liable for violations? A standard clause in the project contract saves discussions later.

How we handle it at Aiara

Transparency is our business — so we disclose how we handle it ourselves: the illustrations on our website and in this blog are partly created with AI tools. They are stylised graphics without real-looking persons — and therefore not subject to the labelling obligation. Our texts are editorially reviewed and accounted for. And our core products — cookie banner, privacy policy, legal notice — are generated rule-based from your input, not by an AI: the same input always produces the same legally vetted result.

The AI Act and cookie law are two different construction sites — but they belong to the same question: is your website legally in order? If you are reviewing your compliance anyway, it is worth looking at the whole picture: we cover the cookie banner, privacy policy and legal notice — and the AI check takes half an hour with the checklist above.