EU Issues Article 50 AI Guidelines on Labelling Synthetic Content

Article 50 requires disclosure in three broad situations: when people interact with AI, when emotion-recognition or biometric-categorisation systems are used on them, and when an AI system has generated or manipulated certain content.

The Commission said the rules respond to the growing difficulty of telling synthetic content from authentic material. It cited risks of misinformation at scale, fraud, impersonation and consumer deception.

Which content must be labelled

Certain AI-generated or manipulated content must carry a clear, visible label and machine-readable marks. The EU has created a set of icons that can be used for this purpose.

The labelling duty covers images, audio and video that resemble real people, objects, places, entities or events, commonly called deepfakes. It also covers text published to inform the public on matters of public interest where no human has reviewed it or exercised editorial control.

Who is affected

Obligations depend on a company's role in the AI chain, Morgan Lewis said. A provider is the organization that developed an AI system. A deployer is the organization that uses it.

Advertisers outside the EU are also in scope. Trade outlet reports say they face fines of up to 3% for unlabelled AI deepfake ads.

The Commission has also confirmed that a transparency Code of Practice is adequate and has published the final version of its guidelines, according to law firm Faegre Drinker Biddle & Reath.

What comes next

A further set of dates arrives on December 2, 2026. Generative AI systems already on the EU market must then comply with Article 50(2) marking, the machine-readable marking duty. New bans on non-consensual intimate material and child abuse material also apply from that date. Both were fixed by Regulation (EU) 2026/1744, the Digital Omnibus on AI.