New EU AI rules take effect: mandatory transparency and content labeling

Starting today, new transparency rules for the use of artificial intelligence (AI) systems come into force across the European Union, as предусмотрed in the European AI Act. They require providers and operators to inform users when they interact with AI and to label content created or modified using such technologies.

The AI Act, which entered into force on August 1, 2024, is the world’s first comprehensive legal framework regulating this field. It introduces common rules for all EU member states based on risk assessment, aiming to ensure that AI development and use are safe, transparent, and aligned with fundamental rights.

Under the new requirements, systems that communicate directly with people, including chatbots, must clearly inform users that they are interacting with artificial intelligence. An exception is allowed only when this is obvious from the nature of the service.

Special attention is also given to so-called "deepfakes". All images, audio, and video created or altered using AI must be clearly labeled. This requirement also applies to text content of public interest generated with AI.

The regulation provides certain exceptions, including materials used for criminal justice purposes, as well as artistic, creative, and satirical works. Labeling is not required when the content has undergone human editorial review and responsibility is assumed by a specific person or organization.

The only delay concerns the requirement for machine-readable labeling or digital watermarking of AI-generated content, which will take effect on December 2, 2026.

Enforcement of the regulation is carried out jointly by the European Commission and national supervisory authorities. At the EU level, oversight of general-purpose models is entrusted to the European Commission’s AI Office.

In Bulgaria, the Communications Regulation Commission has been designated as the national coordinator, participating in the European supervisory system and coordinating implementation at the national level.

The law classifies AI systems into four categories based on risk. Minimal-risk systems, such as spam filters and video games, are not subject to special requirements. Limited-risk systems, including chatbots, must comply with transparency rules. High-risk applications, such as those used in healthcare, transport, or critical infrastructure, face significantly stricter requirements.

The regulation also bans certain practices, including manipulation of human behavior, social scoring of citizens, and some forms of emotion recognition.

In case of violations, national authorities may require corrections, restrict or ban the system, or withdraw it from the market. Fines can reach up to 35 million euros or 7% of global annual turnover. For providers of general-purpose models, the European Commission may impose fines of up to 15 million euros or 3% of turnover, whichever is higher.

Next steps include the introduction of mandatory machine-readable labeling from December 2, 2026, the creation of at least one national regulatory sandbox in each member state by August 2, 2027, and the gradual implementation of rules for high-risk systems.