The EU establishes the rules for the labeling of content generated with AI

Generative AI tools enable the creation of synthetic content —text, images, audio, video— with a level of sophistication that makes it increasingly difficult to distinguish the authentic from the artificial. Without delving deeply into the complex issues of intellectual property that this raises —both regarding the training of algorithms and the attribution of authorship to the generated content—, there is an even greater risk: the impact that the proliferation of this type of content has on our democracies; the potential for blackmail of public figures or anonymous individuals, which can arise from the production of fake content that appears authentic; manipulation and the potential to alter consumer behaviour; in short, the risk is systemic for the health of modern social and democratic life.

If we, as citizens, are unable to distinguish reality from fiction, we run the risk of being deceived, manipulated and, ultimately, making decisions —including how we vote— based on false information. This is not a hypothetical scenario. Deepfakes of political figures and automated disinformation campaigns have already disrupted electoral processes around the world.

Social networks and the digital environment amplify this problem exponentially. Fake content can go viral in a matter of minutes, reaching millions of people before any fact-checking mechanism can act. The informational ecosystem becomes overwhelmed, and it is increasingly difficult for citizens to distinguish between what is real and what is manufactured.

Aware of these risks, the European Union has been deploying a regulatory framework to try to mitigate them. The Digital Services Act (DSA) has already introduced specific obligations for large platforms, which must conduct assessments of systemic risks —including those related to democratic processes and public discourse— to establish mechanisms to report illegal content and remove it in a timely manner. Furthermore, the Code of Practice on Disinformation, which is part of the DSA framework, requires that platforms identify and label manipulated content.

But this is not enough. We need, as citizens, to be able to distinguish when content is genuine and when it has been generated or manipulated by a machine. This is the only way to preserve our critical thinking and make informed decisions in a democratic society.

The EU, fully aware of this need, has taken a decisive step forward with the European Artificial Intelligence Regulation (Regulation (EU) 2024/1689). Article 50 establishes a series of transparency obligations that apply, regardless of the risk level of the system, in four key scenarios: when an AI system interacts directly with people; when it generates synthetic content (audio, images, video or text); when it is used for emotion recognition or biometric categorization; and when it creates deepfakes or publishes texts on matters of public interest. These obligations apply, with specific provisions in each case, to both providers and those responsible for deploying these tools.

Essentially, providers must design their systems in such a way that users are aware that they are interacting with an AI, and they must label the generated content in a machine-readable format, to ensure it can be detected. Implementers, for their part, must disclose when content constitutes a deepfake or when a text on matters of public interest has been generated by an AI without human review or editorial oversight. These obligations come into effect from August 2, 2026, with a grace period until December 2026 for systems already on the market, in accordance with the AI Omnibus.

To facilitate compliance with these transparency obligations, the European Commission has supplemented these provisions with two key instruments. First, the guidelines on interpreting Article 50, published in July 2026, which clarify the scope, definitions, and applicable exceptions. Secondly, the Code of Good Practice on the labeling and tagging of AI-generated content, published in June 2026. This code, which is voluntary, establishes specific measures such as the implementation of a multi-level labeling approach —that combines digitally signed metadata and imperceptible watermarks— and the use of standardized EU icons to visually label AI-generated content. The Commission and the AI Committee have confirmed that the code is an adequate tool to demonstrate compliance with transparency obligations.

The message from the EU is clear: transparency about the origin of synthetic content is not optional. It is a necessary condition for the integrity of the informational ecosystem and, ultimately, for the health of our democracies and our people. It remains to be seen whether the interpretation and implementation of the rules will hinder the production of content and whether the technical mechanisms for marking and labeling will be robust enough to keep pace with the rapid evolution of generative AI, but the framework is already established. The rules of the game are already on the table.

Access the full article published in El Confidencial.