Guide
Article 50 marking obligations, in catalog terms
Article 50 is the EU AI Act’s transparency chapter. Two duties get mixed up in buyer conversations: machine-readable marking of synthetic outputs (Article 50(2), on providers) and visible labeling of deepfakes and some AI text (Article 50(4), on deployers who publish). This page maps those duties to tools already in this directory. It is not a legal opinion and does not certify compliance.
What ‘marking’ usually means (Article 50(2))
Providers of generative systems must ensure outputs are marked in a machine-readable format and detectable as artificially generated, so far as technically feasible. In this catalog that work sits in watermarking (for example Google SynthID, IMATAG) and cryptographic provenance (C2PA Content Credentials).
This site’s free checker looks for Content Credentials / C2PA-style bytes in a file or public URL. It does not detect SynthID-class watermarks in pixels or audio, does not validate signatures, and is not a compliance determination.
What ‘labeling’ usually means (Article 50(4))
Deployers who publish AI-generated or manipulated media of real people, places, or events must disclose that visibly. AI-generated text published to inform the public on matters of public interest has a related labeling duty, with an editorial-review exception described in the statute — get counsel for your facts.
Detection products in the directory (Hive AI Detection, Reality Defender, and others) help platforms notice unlabeled synthetic media. They are not a substitute for a visible label, and they are not this checker.
A practical order of operations
Check a sample output for leftover C2PA / Content Credentials signals. If none appear, you have not ruled out an invisible watermark — compare SynthID vs C2PA next.
If you generate the content, pick a marking family from the catalog and keep it through delivery (CDNs often strip metadata). If you publish other people’s media, pair visible labels with detection rather than assuming files arrive marked.
Tools from this catalog
Cards use the same one-liners as the directory. We do not add capabilities that are not on the listing.
Related
Informational directory content, not legal advice. Last substantive review: 19 September 2026.