The EU has started labeling content created by artificial intelligence

From August 2, the European Union has begun to enforce transparency requirements for artificial intelligence. Users must be warned that they are interacting with AI, and content created or substantially altered by algorithms must be provided with special labels.

The new rules are set out in Article 50 of the European AI Act. They apply to providers of AI systems and companies, public bodies, or other organizations that use such technologies.

Developers must configure systems so that a person understands when they are communicating with a chatbot or other AI. An exception is possible when the artificial nature of the system is obvious from the context.

Texts, images, audio, and video created or substantially altered by artificial intelligence must receive a machine-readable mark. This will allow platforms and special tools to determine the origin of the material. Ordinary technical processing that does not change the meaning of the source data does not fall under this requirement.

Deepfakes – realistic images, audio, or video that may create a false impression of authenticity – must be clearly labeled separately. For artistic, satirical, and fictional works, a more flexible form of warning is provided that should not interfere with the perception of the work.

Labeling is also mandatory for AI-generated texts on matters of public interest, unless they are published with substantive human review and editorial control. Materials for which the editorial team is responsible and that have undergone full checking do not require additional labeling. Simple spelling checks are not considered editorial control.

Users must also be notified about the use of emotion recognition systems and biometric classification systems.

Companies that breach the transparency requirements may face a fine of up to 15 million euros or up to 3% of their global annual turnover, whichever is higher. The principle of proportionality of punishment is established for small and medium-sized enterprises.

For some generative systems that appeared on the market before August 2, 2026, a transitional period for technical labeling is valid up to December 2. Content published before the rules came into effect does not need to be retrospectively labeled. 

Based on: European Commission