Article 50(2) · due 2 December 2026

Mark your AI-generated images

Article 50(2) requires AI-generated content to be marked in a machine-readable format — a visible caption on its own does not satisfy it. In our scan of 883 EU sites, only 0.4% carried any such marking. This writes the IPTC provenance straight into your file. It runs in your browser; the image is never uploaded.

Drop a PNG or JPEG here
or click to choose · nothing leaves this page

Also do these two things

Machine-readable marking is the part with a deadline, but it is not the whole duty. Article 50(4) separately requires deployers to disclose deepfakes and AI-generated text published on matters of public interest.

1. A visible label, in the reader's language

2. Structured data on the page

So crawlers and platforms can read the claim without downloading the asset. Paste into the page that publishes the image.

What this does and does not do

It writes IPTC digitalSourceType into an XMP packet inside the file — an iTXt chunk for PNG, an APP1 segment for JPEG. That is the vocabulary the AI Act's machine-readable requirement is generally read against, and the same one Adobe, Google and the major agencies use.

It does not sign anything. A full C2PA Content Credential is cryptographically signed and tamper-evident. XMP is a plain assertion — anyone can strip or forge it. For journalism or evidentiary use you want C2PA proper; for a marketing image, IPTC provenance is the proportionate step.

Metadata gets stripped. Most social platforms and many CDNs discard XMP on re-encode. Keep the marking on the original and on what you serve from your own domain, and do not assume it survives a trip through someone else's pipeline.

Information and tooling, not legal advice. ClearLabel is not a law firm.

Is anything else on your site exposed?

The free scan checks whether your site discloses AI chat under Article 50(1), and reports what it can and cannot see. Takes about ten seconds.

Scan your site