Article 50(2): marking generated output
Article 50(2) binds providers of systems that generate synthetic audio, image, video or text: output has to be marked in a machine-readable format and detectable as artificially generated or manipulated, with solutions that are effective, interoperable, robust and reliable as far as is technically feasible. From 1 source page, read 2026-09-12; no maker pays to be listed.
| Point | What the paragraph states |
|---|---|
| Who it binds | Providers, including of general-purpose AI systems |
| What it asks for | Machine-readable marking, detectable as generated or manipulated |
| Quality bar | Effective, interoperable, robust and reliable, as far as technically feasible |
| Carve-outs | Assistive standard editing; systems that do not substantially alter input |
Inclusion rule. One page per numbered paragraph of Article 50, with the same four points on each. A point the paragraph does not address is recorded as not addressed rather than filled from a recital or a summary. Order. Fixed point order, identical on every paragraph page.
1Four adjectives, and only three are a vendor's to deliver
Effective, robust and reliable are properties a single company can engineer and test on its own. Interoperable is not: a mark is interoperable only when parties other than its author can read it, which makes it a property of an ecosystem. A vendor can build the first three alone and cannot claim the fourth alone.
That is the reason this record keeps a separate row for what each generator publishes about machine-readable marking, and why nine of those rows hold no such mark. Seven entries describe a signal or a manifest inside the file. The rest describe what a viewer sees, which is the subject of a different paragraph.
2The technical-feasibility qualifier is doing real work
The obligation is bounded by what is technically feasible, taking into account the specificities and limitations of different types of content and the generally acknowledged state of the art. A duty written that way moves as the field moves, which means a position that is defensible in one year is not automatically defensible later.
Nothing in the paragraph names a standard, a format or a registry. Anyone looking for the specific field to write is looking for something the text does not supply, and the codes of practice mentioned further down the article are the mechanism the regulation points at instead.
- Article 50(2)Providers must mark synthetic audio, image, video and text in a machine-readable formatincluding general-purpose AI systems
- Article 50(2)Technical solutions must be effective, interoperable, robust and reliable as far as technically feasiblequalifier on the marking duty
- Article 50(2)Machine-readable marking does not apply to assistive standard editing that does not substantially alter the inputsecond exemption
3A provider's own account
Against the duty in “Article 50(2)”, SceneMixer's published pages describe what its pipeline produces:
It becomes a marked line — inner voice or narration — so it is heard while the picture shows something worth watching, rather than becoming a shot of somebody thinking.
SceneMixer, AI Dialogue Generator — “What happens to inner monologue from a novel?” (read 2026-10-13)
Dubbing replaces the audio of a finished clip; the picture was generated for a different line, so mouth shapes, pauses and shot length all belong to the original language. Native-language generation gives the video model the line in the target language before it renders the shot, so the performance, the pacing and the cut points are built for that language.
SceneMixer, Native-Language Dialogue vs. Dubbing in AI Video (2026): Why the Cast Should Speak the Line (read 2026-10-13)
4Sources
Wording read from AI Act Explorer, Article 50 on 2026-09-12. The whole article in one place is on the Article 50 page; who each duty binds is on duties by role. Next to it: Article 50(1), Article 50(3).