AI Supervision
What is the AI Act and why it affects you
Today, AI decides on matters that affect you directly. From advertisements to loan applications. The AI Act gives you confidence that these decisions will be fair, transparent and under control. We will show you your rights.
What the AI Act means to you
Ban on manipulation and invisible surveillance
The right to know that you are talking to AI
AI transparency and protection against deepfakes
Right to an explanation of AI decisions that fundamentally affect you
Right to review AI decisions that fundamentally affect you
What AI must not do today
Prohibited practices already apply today, even in the absence of a national law. The main role of the CTU is now to educate and explain what the rules mean in practice.
PROHIBITED
Manipulation and misuse
Subliminal manipulation and exploitation of vulnerabilities, such as age or health.
PROHIBITED
Monitoring and assessment
Social scoring, predictive policing systems and emotion recognition at school or at work.
PROHIBITED
Biometrics
Blanket scraping of faces from the internet and illegal biometric identification or categorisation.
Videos #AIRULES
Video
Video
Video
Do you run a business or manage an office?
Visit CTU webpageFAQ
How exactly does the AI Act regulate deepfake content?
The AI Act does not prohibit the creation of deepfake content as such. It focuses mainly on transparency – so that users know that they are working with content produced or edited with the help of artificial intelligence. Those rules also include a requirement for the technical labelling of such content, which may facilitate its later identification.
At the same time, any creation or dissemination of deepfake content is not automatically allowed. Harmful uses of this technology may be punishable under other laws, such as criminal law, the protection of the personality or the protection of personal data.
The fact that the AI Act will prohibit, as of 2 December 2026, certain AI systems enabling the creation of realistic intimate materials that exploit the identity of a real person without their consent can also be seen as an important step. The aim is therefore not to prohibit technology as such, as it can have a number of legitimate uses in, for example, education, science, medicine or the arts, but to limit its harmful use.
Who is concerned by the AI Act?
Operator of AI system
First and foremost, they must ensure that the so-called high-risk AI systems they develop are safe and reliable. To this end, they must comply with requirements covering areas such as risk management, transparency for users, human oversight, data handling, technical documentation, logging and the reporting of serious incidents. Further obligations relate, for example, to the labelling of ‘synthetic content’ generated by AI systems and the registration of AI systems in the EU database.
Importer and distributor of AI system
Each has specific obligations, largely of an administrative nature, depending on their position in the supply chain.
Organisations using AI systems (deployers)
In particular, they are required to use high-risk AI systems in accordance with the instructions of use and under the supervision of a qualified person, to monitor their operation or to report serious incidents. In selected cases, they must, for example, assess the fundamental rights impacts of the use of the AI system, register its use in an EU database, or inform affected employees and other persons that they will be exposed to the AI system.
What rules apply from 2 August 2026?
The obligation to provide information about communication with an AI system - Article 50(1) AI Act requires providers of AI systems to ensure that natural persons are adequately informed that they are interacting with a chatbot, voicebot or other system designed to interact directly with natural persons in an appropriate manner.
The marking and labelling of AI-generated content – the CTU recalls that, according to Article 50(2), the obligation for providers of AI systems to ensure that the outputs of such systems are properly marked in a machine-readable format and are thus detectable applies to all audio, image, video or text content, subject to the specified exceptions. By contrast, the deployer’s obligation under Article 50(4) only applies to deep fakes and texts published for the purpose of informing the public on matters of public interest. Consequently, where image, audio or video content is generated or significantly modified by a deployer using AI in a way that may give the impression of authenticity, the deployer shall inform the public that this content has been generated or modified with the assistance of artificial intelligence. The same applies to text published for the purpose of informing the public about matters of public interest, which has been generated or manipulated by an AI system. In that regard, the CTU points out that, under Article 50(5) AI Act, clear and distinguishable means of labelling must be used. In practice, for example, labels or icons of the type ‘AI generated’ for fully generated content or ‘AI modified’ for AI-modified content can be used. Examples of good labelling practices can be found in the Code of Practice on Transparency of AI-Generated Content.
Emotion recognition and biometric categorisation – Under Article 50(3), deployers using AI systems for emotion recognition or biometric categorisation of natural persons shall duly inform each natural person concerned of their use. The CTU also recalls that AI Act provides for a number of exceptions to the above-mentioned rules, such as for purposes related to the detection of crime, etc., and recommends the entities concerned to assess whether a given activity falls under any of these exceptions.
Does the obligation to mark and label AI content also apply to media?
As a general rule, under the AI Act, the media are not exempt; they are entities like any others, are subject to the AI Act, and must act in accordance with it. However, we would like to draw your attention to the exception to the obligation to state that artificial intelligence has been used to generate or modify a text to inform the public on matters of public interest in cases where such a text has subsequently undergone a human review or editorial control process and where the publication is under the editorial responsibility of a (specific) natural or legal person.