What the rule is
Starting 2 August 2026, Article 50 of the EU AI Act requires visible disclosure of certain AI-generated content. The rules address two situations:
Deepfakes — AI-generated image, video or audio that resembles real people, places or events convincingly enough to be mistaken for authentic
AI-generated text on matters of public interest — health, safety, elections — where no human has reviewed it and taken editorial responsibility
Most everyday AI-generated content (ad copy, product descriptions, social posts, newsletters) does not require labeling. The obligation only kicks in when the content could mislead someone about whether it's real.
To support compliance, the European Commission published a voluntary Code of Practice on Transparency of AI-Generated Content, signed by roughly 190 organisations by end of July 2026, alongside an official set of free-to-use EU icons for labeling.
Who this affects
The rules apply to anyone using AI in a professional capacity — whether you're a brand, an agency, or an individual creator. The legal term is "deployer": if you use an AI system to create or publish content professionally, you're in scope. Purely private, non-professional use is exempt.
If you work as an agency publishing on behalf of a client, the responsibility can sit with both of you at once.
What actually needs a label — and what doesn't
This is the most practical part of the regulation. Work out which column your content falls into first — everything else follows from that.
Label it:
Photorealistic people: a synthetic influencer, model, or spokesperson, even if fully invented
Face swaps of a real person or employee — consent does not remove the duty
A voice imitating a specific real person, even in a short clip
Photorealistic scenes in a real setting that could pass as a record of something that happened
AI product or packaging imagery that could mislead about how the product actually looks or performs
AI text on public-interest topics: news, politics — health, consumer safety and sustainability count as public interest even in a commercial context
No label needed:
Illustrations, vectors, animations, stylised graphics — anything that does not pretend to be a record of reality
Obvious fantasy: talking mice, a tomato with legs, a dragon driving a car
A real product shot against an AI background, as long as the product itself is truthful
Standard aesthetic edits: colour correction, denoise, lighting, crop, background replacement, rescaling
Ad copy, product descriptions, ordinary social posts, newsletters, promo text — not matters of public interest
AI text a human genuinely rewrote, fact-checked and took responsibility for — a typo check does not qualify
Three exceptions worth noting:
Artistic, satirical or fictional work — disclosure can be lighter-touch, but never skipped entirely. Advertising almost never qualifies. A deepfake of a real person is never exempt just because it's satire.
Law enforcement use — where legally authorised for detecting or investigating crime.
Human-reviewed text — where a named person or organisation holds genuine editorial responsibility.
How platforms handle AI labels today
Some platforms already read AI metadata (C2PA Content Credentials) from your files at upload and display their own label automatically. Based on our own field testing in August 2026:
Facebook — displays "AI content" next to the author name. Works on single images and gallery posts.
LinkedIn — displays a Content Credentials panel behind a small "cr" icon. Visible to any viewer, not just the post owner.
Instagram — displays "AI info" on single images, but does not show it on carousel/multi-image posts.
Pinterest — displays a visible "AI modified" banner directly on the pin.
Important caveats from our testing:
Detection depends on which AI tool created the file — the same type of content produced by different tools can give different results
Video is not reliably detected on any platform we tested
Kontentino preserves C2PA metadata — verified by uploading files to our library and downloading them via Download Assets
You can check your own files before publishing at verify.contentcredentials.org — the official, free verification tool by the Content Authenticity Initiative. Just drag your file in.
Because automatic platform labels are inconsistent, we recommend not relying on them as your only disclosure method. A visible label baked into the image or video itself is the only approach that survives every path: reshare, screenshot, download, and format conversion.
How to mark AI-generated content in Kontentino today
Option 1: Automatic platform detection (when it works)
If your AI tool embeds C2PA metadata (Adobe Firefly, ChatGPT/DALL-E and others do this), Kontentino preserves it during publishing. Some platforms will then display their own AI label automatically — see the section above for which platforms do what.
Option 2: Visual label via Watermark (recommended as a reliable fallback)
You can add a visible AI label directly into the image using Kontentino's watermark feature:
Open your post and click Edit.
Go to the Watermark option and select Add watermark.
Upload an AI label image as your watermark.
You can use the official EU icons (download below) or a brand-consistent equivalent, as long as it's equally visible and clear.
According to the EU Commission's guidance: "The icon must be visible when content is reshared or downloaded." A label baked into the pixels is the only method that guarantees this.
Coming soon: API-level disclosure
The platforms are beginning to offer API parameters for AI disclosure (Instagram added is_ai_generated in June 2026). We are tracking these developments and will implement them as they become available across platforms.
Download EU icon set
The European Commission provides a free icon set for labeling AI-generated content. No attribution required. Three variants are available:
AI — basic icon, paired with your own text label
AI Generated — for fully AI-generated content
AI Modified — for real content that AI turned into a deepfake
Available in black and white, with and without transparent background.
Official source: EU icons for labelling AI-generated content
Where to find more
Download the full guide
We've put together a detailed slide deck covering everything in this article, including real examples, our field test results, and practical placement guidelines by format.