EU Article 50 Defines AI Disclosure Exceptions

Search Engine Journal characterises Article 50 as containing four exemptions relevant to publishers and AI content. The underlying legal provisions and Commission guidance separate several distinct situations rather than creating a single blanket exemption for AI-assisted publishing.

• •Standard editing and assistive functions: Article 50's machine-readable marking obligation does not apply where an AI system performs an assistive function for standard editing, or where it does not substantially alter input data or its semantic meaning. This provision concerns providers' system-design and output-marking duty.
• •Human-reviewed public-interest text: Under the Commission guidance, publication disclosure for AI-generated text on matters of public interest does not apply when there has been human review or editorial control and someone holds editorial responsibility.
• •Artistic, creative, satirical, or fictional deepfakes: For deepfake content in evidently artistic, creative, satirical, fictional, or analogous works, Article 50 requires disclosure in a manner that does not hamper the work's display or enjoyment. This is a presentation qualification, not necessarily a complete removal of disclosure duties.
• •Legally authorised uses: Article 50 provides an exception in circumstances where use is authorised by EU or national law to detect, prevent, investigate, or prosecute criminal offences.

The Commission's FAQ separately notes that providers of generative AI systems must use effective, reliable, robust, and interoperable machine-readable marks to make AI-generated or manipulated outputs detectable. It also requires systems that directly interact with people, including chatbots, agents, and avatars, to inform users that they are interacting with AI.

The publisher-specific text rule turns on more than whether a draft began with a generative model. According to the Commission guidance, the relevant factors are whether the publication concerns matters of public interest, whether humans exercised review or editorial control, and whether an identifiable natural or legal person retains editorial responsibility.

There is also an implementation caveat. The AI Act Service Desk warns that its displayed Article 50 text has not yet been updated to reflect Digital Omnibus amendments. Separately, a May 2026 Clifford Chance analysis described a possible transitional period for certain Article 50(2) machine-readable marking requirements affecting general-purpose AI systems placed on the market before August 2. Organisations assessing compliance should therefore distinguish the Commission's current guidance from the consolidated legal text and any applicable transitional provisions.

The Commission frames the obligations as measures intended to help people identify AI interactions and AI-generated content, calibrate reliance on it, and reduce risks including misinformation, fraud, impersonation, and consumer deception. Article 50 guidance also distinguishes which controls apply to the model provider, the deployer, and the editorial organisation that releases the final content.