What is the EU AI Act disclosure rule?

What Article 50 actually requires

Providers of AI systems that generate synthetic audio, image, video or text content must ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. This is a technical obligation on the tool maker, and a visible watermark alone does not satisfy it. It is meant to let downstream platforms and detection tools identify AI content automatically.

Deployers, meaning the businesses or individuals actually using an AI system to generate or manipulate content, face a separate disclosure duty. If they generate or manipulate image, audio or video content that constitutes a deepfake, meaning it depicts real people, places or events in a way that appears authentic, they must disclose that the content is artificially generated or manipulated.

Deepfakes get a specific carve-out

The deepfake disclosure duty applies whether the deepfake is used for entertainment, satire, art, or anything else, though the AI Act allows the disclosure to be done in a way that does not hamper the display or enjoyment of the work, such as a discreet label rather than an intrusive watermark plastered across the content.

Text generated or manipulated by AI that informs the public on matters of public interest carries its own disclosure obligation too, unless the content has undergone human review and a natural or legal person holds editorial responsibility for it, which is a carve-out aimed at newsrooms using AI-assisted drafting with human editors in the loop.

Machine-readable marking versus visible labels

Machine-readable marking, sometimes implemented through standards like C2PA content credentials, embeds metadata into the file itself so platforms and detection systems can identify AI-generated content programmatically, even if a human viewer sees nothing unusual. This differs from a visible on-screen disclosure aimed at a human audience, and Article 50 contemplates both depending on the obligation and the actor involved.

Timeline and enforcement

The EU AI Act entered into force in August 2024, with obligations phasing in over several years by risk category. The transparency obligations under Article 50, covering marking and disclosure, apply from 2 August 2026, and national market surveillance authorities in each member state are responsible for enforcement, with penalties tied to the Act's broader fine structure.