We generate marketing images of products, not people. Deepfake?

Maybe not under 50(4). Still a 50(2) marking problem for the provider. Put a visible “AI-generated image” label if you want one disclosure that covers both duties; the labeling snippet does that.

Who discloses — the model vendor or the brand that posted the video?

50(4) names the deployer of the system that generated or manipulated the content. The brand that ran the generator and posted the video is the usual deployer. The model vendor is the usual provider (50(2)).

Deepfake disclosure under EU AI Act Article 50(4)

Article 50(4) first subparagraph is a deployer duty: if you generate or manipulate image, audio, or video that constitutes a deepfake, you disclose that the content was artificially generated or manipulated. The duty applied on 2 August 2026. It is not covered by the Article 111(4) marking transition.

Educational resource only — not legal advice. scm.cc publishes a readiness report and engineering gap analysis for AI features. It does not give legal advice, represent you before a regulator, or state that a product is ready to ship under any law. This page is educational product guidance, not legal advice. It is not a readiness report, not an engineering gap analysis of your systems, and not a determination that any product meets a legal duty. Confirm current official text before you ship.

Last checked against official sources: 2026-10-11.

What counts, in the official words

The article uses “deep fake” for image, audio, or video content generated or manipulated by an AI system. The Commission guidelines add examples and edge cases. A face-swap video of a real person and a synthetic voice clone of a real person are the core cases. A fully invented landscape still needs 50(2) marking from the provider; whether it is a 50(4) deepfake is a guidelines question — read them before you skip the visible label.

The artistic-work exception

Where the content forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme, the duty shrinks to disclosing the existence of such generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work. A film credit, a hover label, or an info panel can be enough; a 30% opaque banner across the frame is more than the article asks for.

Law-enforcement carve-out

The duty does not apply where the use is authorised by law to detect, prevent, investigate or prosecute a criminal offence. That is not a “trust and safety team” exception. It is a legal-authorisation exception.

Questions teams ask

We generate marketing images of products, not people. Deepfake?: Maybe not under 50(4). Still a 50(2) marking problem for the provider. Put a visible “AI-generated image” label if you want one disclosure that covers both duties; the labeling snippet does that.
Who discloses — the model vendor or the brand that posted the video?: 50(4) names the deployer of the system that generated or manipulated the content. The brand that ran the generator and posted the video is the usual deployer. The model vendor is the usual provider (50(2)).

Back to the AI law hub

Primary sources

Official text wins. Last checked 2026-10-11. This page is educational product guidance, not legal advice. It is not a readiness report, not an engineering gap analysis of your systems, and not a determination that any product meets a legal duty. Confirm current official text before you ship.