From August 2, the AI Act requires AI systems to be identified. Companies must mark synthetic content and notify users during interactions.
The 2nd of August marked a turning point in the relationship between companies and artificial intelligence. That day, the transparency obligations of Article 50 of the European Artificial Intelligence Regulation, known as the AI Act, came into effect, a set of rules that affects any system that interacts with people or generates synthetic content.
The regulation is not limited to high-risk systems. It also covers everyday uses such as customer service, marketing, communication, or content generation. For many companies, the question is no longer what AI can do, but how to know when it is acting and how to identify what it has produced.
The first obligation affects systems that converse directly with people. Providers must design them so that the user knows they are speaking with a machine, unless it is evident from the context. With synthetic voices, memory, and agents executing actions, the boundary between the human and the artificial blurs, and the regulation seeks to prevent someone from believing they are speaking with a person when they are not.
The second obligation is more technical and affects generators of audio, image, video, or text. Their outputs must be marked in a machine-readable format, through metadata, watermarks, or content credentials, so they can be detected as artificial. The difference is key: labeling informs a person, while digital marking also allows a machine to be informed.
Deepfakes receive special attention. Images, audio, or videos that may be perceived as authentic must reveal their artificial origin, although the regulation allows for more flexible treatment for artistic, creative, satirical, or fictional works.
Not all AI-generated content needs to be labeled. Article 50 establishes exceptions, for example, when there is human review or editorial control and a person or entity assumes responsibility. It is not the same to correct a text with AI as it is to generate it entirely, nor to adjust a photo as it is to create a non-existent person.
There is a limited transitional period. Until December 2, 2026, systems introduced to the market before August 2 are exempt from the marking and detection obligations of Article 50.2. This is not a general extension, and there is no need to retroactively label content published before that date.
For companies, compliance is not just about putting a notice on the website. They must identify the AI systems used, classify them, and determine which obligations apply. The Navarra Chamber of Commerce, through its Acelera Pyme Office, has organised meetings with digitalising companies in the IRIS Digital Innovation Hub, with the Government of Navarra and NAIR Center, to translate the regulation into real processes.
Transparency must be incorporated from the design stage, as is already done with privacy or security. Before implementing an AI, it is advisable to know what system is being used, who the provider is, what data is processed, what human oversight exists, and who assumes responsibility.
The harder it is to distinguish the artificial from the human, the more valuable it will be to certify its origin.
August 2 is not just a date for the AI Act. It marks the beginning of a new rule for digital coexistence: in certain cases, we have the right to know when there is artificial intelligence behind a conversation, an image, or information.
You may also like
