It was supposed to be one of the main tools to make AI-produced content recognizable, but the labeling system outlined by the European Commission arrives at the final version with some modifications that narrow its application.
Between pressures from the advertising industry and the fear of excessive labeling, the boundary between transparency and simplification returns to the center of the deepfake debate.
WHAT THE NEW GUIDELINES PROVIDE
The European Commission’s guidelines clarify how providers and users of AI systems must apply the transparency obligations set out in Article 50 of the AI Act, which will come into force next August 2. The document complements the Code of Practice on Transparency of AI-Generated Content and provides guidance on the meaning of concepts such as interactive systems, synthetic content, deepfakes, and AI-generated texts on matters of public interest.
Article 50 establishes that providers must design systems “in such a way that the individuals concerned are informed that they are interacting with an AI system,” except in cases where this circumstance is obvious to a reasonably informed and attentive user. Furthermore, systems that generate synthetic audio, images, video, or text content must produce results “marked in a machine-readable format and detectable as artificially generated or manipulated.”
The rules also concern users of AI systems. Those employing emotion recognition or biometric categorization tools must inform the people involved, while those distributing deepfakes or artificially generated texts on issues of public interest must declare their artificial origin, except for certain exceptions provided by the regulation.
OPENINGS ON MODIFIED CONTENT
One of the most debated points concerns the scope of content that must be labeled as generated or manipulated by AI. In the final version of the guidelines, the Commission has expanded some exemptions compared to the previous draft.
For example, notes Euractiv, replacing a background in an image for “clearly aesthetic purposes” may not require a label, even when the modification is significant. According to the guidelines, this type of intervention would generally have a limited impact on the likelihood that a person considers the image authentic.
The previous draft version instead provided a narrower exemption, limited to extending or modifying background details. The new wording therefore expands the scope for graphic interventions made with AI tools without the obligation to report them.
PRESSURES FROM THE ADVERTISING SECTOR
The relaxation of some conditions, states Euractiv, comes after requests from advertising operators, who argued the need to avoid excessive spread of AI labels. According to advertisers, the use of AI-based editing tools has become common, and too many warnings risk producing a habituation effect among users, reducing the effectiveness of the transparency indications themselves.
The Commission has confirmed that one of the objectives of the guidelines is to avoid what is called “labelling fatigue,” that is, a condition in which the continuous presence of labels leads people to pay less attention to warnings. However, the principle remains that advertising must not create a false representation of the real characteristics of a product.
LESS CERTAINTY ON WARNINGS IN VIDEO CONTENT
Another element modified compared to the initial draft concerns broadcasts and streams containing deepfakes. The previous version required that warnings about the presence of artificially modified content be visible even in later stages of the broadcast or permanently, excluding that a simple notice in the end credits could be sufficient.
The final version instead introduces less rigid criteria, establishing that such indications must be provided “where possible” and “as appropriate,” without explicitly requiring the warning to be maintained throughout the entire content or in the final credits.
This change directly affects the number of labels users might encounter in practice: not all AI-modified content will necessarily be accompanied by a visible notice, especially in cases where the manipulation is not considered sufficiently relevant.
APPLICATION OF THE RULES
The new provisions will be enforced through the supervision system provided by the AI Act, involving national market surveillance authorities, the European Artificial Intelligence Board for systems under its supervision, and the European Data Protection Supervisor in cases concerning European Union institutions.
Compliance with the obligations can also be demonstrated through adherence to the Code of Practice on Transparency of AI-Generated Content. For entities choosing not to adhere, it will still be necessary to demonstrate the adoption of alternative solutions considered equivalent to ensure the marking and identification of artificially generated or manipulated content.
