The AI Act, the regulation with which the European Union regulates the development and use of artificial intelligence, will enter its most important phase from 2 August 2026. The innovation that will have the most obvious impact on daily life concerns transparency: from now on users must be able to clearly understand whether they are interacting with a person or with an AI-based system, and the same principle will also apply to artificially created texts, images, audio and videos. The goal is not to limit the use of AI, let’s be clear, but to make sure that everyone knows whether they are interacting with AI technology or not.
What changes with the AI Act: the application phases
The AI Act is the European regulation designed to establish common rules on artificial intelligence, introducing obligations and responsibilities for those who develop or use these systems. Its application was planned in several phases. Since February 2025, systems considered to be at unacceptable risk, such as those used for mass surveillance, have been banned. From August of the same year, the provisions relating to governance came into force, i.e. the set of rules that define controls, responsibilities and supervision methods, as well as the system of sanctions. However, 2 August 2026 represents the decisive moment for the rule, as the part of the regulation that involves the highest number of companies, bodies and professionals becomes applicable. The implementation process will then continue in 2027 and 2028 with further obligations provided for by the regulation.
The key principle of the new provisions is contained in Article 50 of the legislation and introduces a very simple right to understand: the user must be able to know whether what he is reading, listening to or observing was created by a person or by an artificial intelligence system. Transparency therefore becomes one of the central elements of European legislation, requiring companies to make the presence of AI explicit when it interacts directly with users.
For this reason, chatbots, virtual assistants and voice systems will have to be designed so that it is immediately evident that there is not a human operator on the other end. Whether it is a customer service, a system for making reservations or a tool that collects initial commercial information, the interaction with a machine must be clearly stated, except in cases where this is already obvious from the context. For many companies this will involve changes to interfaces, messages shown on the screen and voice communications.
Similar obligations will also concern systems that produce artificial content. Texts, photographs, audio and video recordings generated by artificial intelligence must be marked with recognizable and legible marking, within the limits permitted by available technology. A recent European standard, known as Digital Omnibus, however, gives more time to systems already on the market before 2 August 2026: the adaptation can in fact take place until 2 December 2026. Those who develop or integrate these tools will have to verify together with suppliers which identification systems are available and how resistant they are to any subsequent changes.
Particular attention is paid to deepfakes, i.e. content in which images, videos or audio recordings are created or altered by artificial intelligence until they appear authentic. In these cases it will be necessary to declare that the material has been artificially generated or modified, with some exceptions provided for artistic, satirical or creative works. Specific rules also concern texts produced by the AI and published to inform the public on topics of general interest. The obligation to declare the origin only ceases if the content has been subjected to human editorial control and a person takes responsibility for it.
The legislation also introduces obligations for those who use emotion recognition systems or biometric categorization tools. If these tools are used, for example for marketing purposes, the data subjects must be informed and the data processing must comply with European privacy protection rules. This is an area that affects contexts such as workplaces, shops, events, services open to the public and private security.
What happens to those who don’t respect the new rules
The consequences for those who do not respect the new rules can be very serious. Violations can in fact lead to fines of up to 15 million euros or 3% of the annual global turnover, choosing the higher amount, with reduced limits for small and medium-sized businesses. The risk, however, does not only concern fines. Errors generated by artificial intelligence can also have professional consequences, as demonstrated by cases of documents filed with inaccurate information produced by AI and already subject to sanctions. While larger organizations have often already surveyed the tools used and adopted control procedures, many small businesses and professionals have yet to identify all the systems used and develop the skills necessary to manage the risks and responsibilities envisaged by the AI Act.
