Guide · Art. 50(2)

Labeling AI-Generated Content Under Article 50(2)

Last updated: 8 November 2026

AI-generated content under Article 50(2) must carry a machine-readable mark, like C2PA or IPTC metadata or a watermark, added by the provider of the generative tool. Publishers should keep that metadata intact; visible disclosure is a separate Article 50(4) duty for deep fakes and public-interest text. The free Article50.io scan shows the most severe finding on one public page.

Below: what “machine-readable” and “detectable” mean in practice, who the duty applies to, and what machine-readable marking looks like.

Scan for free

What Article 50(2) requires

Providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content must ensure the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. The technical solutions must be effective, interoperable, robust and reliable as far as technically feasible, taking into account the type of content, implementation costs and the generally acknowledged state of the art.

The obligation doesn't apply to AI that performs an assistive function for standard editing, or that doesn't substantially alter the input data or its semantics — a spell-checker or auto-crop isn't in scope.

Timing: Article 50 has applied since August 2, 2026. Under Article 111(4), inserted by the Digital Omnibus on AI, providers of generative systems already on the market before that date have until December 2, 2026 to bring them into line with Article 50(2). Systems placed on the market from August 2, 2026 onward must comply from launch.

Who it applies to

Generative tool makers

Image, video, voice and text generators are providers and must mark what their systems output.

Sites with their own AI features

If your product generates content for users — AI images, summaries, voiceovers — you're the provider of that feature and the marking is yours to build.

Publishers using AI tools

You rely on the tool's marking — so don't strip it — and owe your own disclosures for deep fakes and public-interest text under Article 50(4). See Article 50 for publishers.

What machine-readable marking looks like

Machine-readable marking comes in a few forms, and robust setups layer more than one. Common approaches include embedded provenance metadata such as C2PA Content Credentials, the IPTC DigitalSourceType property, invisible watermarks, and markup that flags AI-generated media on the page itself.

For what Article 50(2) itself requires, marking plus detection, the quality bar and who has to do it, see EU AI Act watermarking: Article 50(2) machine-readable marking.

Image · on-page markup plus a visible label

<figure>
  <img src="/hero.webp" alt="Product lifestyle scene" data-ai-generated="true">
  <figcaption class="ai-label">AI-generated image.</figcaption>
</figure>

Image file · IPTC digital source type (XMP metadata)

<Iptc4xmpExt:DigitalSourceType>
  http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia
</Iptc4xmpExt:DigitalSourceType>

Text · marked container plus a visible disclosure

<article data-ai-generated="true">
  <p class="ai-text-notice">This article is AI-generated.</p>
  <!-- AI-generated article body -->
</article>

Labeling by media type

✔Images — keep C2PA / IPTC metadata through your CMS and CDN; flag AI images in markup
✔Video — keep provenance metadata in the file; add an on-screen or in-player label
✔Audio — use watermarked output where the tool supports it; label synthetic voices in the player
✔Text — mark AI-generated blocks in markup and pair them with a visible disclosure

How Article50.io checks this

The scan looks for concrete, machine-readable AI-generation markers on your public pages — generator metadata from tools like Midjourney, DALL·E, Stable Diffusion, Runway, Synthesia, Sora and Firefly, and explicit data-ai-generated attributes — and flags when nothing visible discloses that content to visitors. It never infers AI generation from how content looks.

For the full set of obligations, see Article 50 of the EU AI Act, explained or the Article 50 compliance checklist.

Frequently Asked Questions

What does "machine-readable marking" mean under Article 50(2)?

It means the AI-generated output carries a signal that software, not just people, can detect — such as embedded provenance metadata, a watermark, or a markup attribute — so the content can be identified as artificially generated or manipulated.

I use a third-party image generator. Who is responsible for marking?

Article 50(2) places the marking obligation on the provider of the generative AI system — the tool maker. Your job as a publisher is to avoid stripping that marking, and to disclose deep fakes and public-interest text under Article 50(4).

Does every AI-assisted edit need to be marked?

No. Article 50(2) doesn't apply where the AI system performs an assistive function for standard editing, or doesn't substantially alter the input data provided by the deployer or its semantics.

Is a visible "AI-generated" caption enough?

A visible caption helps people, but Article 50(2) specifically asks for machine-readable marking. Use both: keep the embedded metadata intact and add a visible label where the content appears.

Check your site automatically

Article50.io is an automated scanner for the Article 50 transparency rules of the EU AI Act. It scans a website for potential Article 50 transparency gaps and generates a report of what the scan found and what it could not check, with remediation guidance, implementation instructions and disclosure wording you can paste in.

The free scan shows your single most severe finding. Scans are capped at 45 seconds — no signup, public pages only.