Does the AI Act ban AI?

What does Article 5 prohibit, in outline?

Article 5 prohibits defined AI practices that are considered to pose unacceptable risks to health, safety and fundamental rights. In outline, as originally enacted and as amended by Regulation (EU) 2026/1744, the prohibitions include manipulative and deceptive practices that materially distort behaviour, exploitation of vulnerabilities, social scoring with defined effects, untargeted scraping of facial images, emotion inference in work and education with defined exceptions, biometric categorisation on sensitive attributes, real time remote biometric identification in publicly accessible spaces for law enforcement with defined exceptions, and predictive policing based solely on profiling, together with the two new prohibitions from 2 December 2026 on AI systems generating or manipulating non consensual intimate material and material within Directive 2011/93/EU on child sexual abuse material, subject to the consent and defence conditions in the Article. That outline is not a substitute for the Article text, which sets out each prohibition with its conditions, exceptions, definitions and application dates, and which must be read in full for any system that raises a prohibition question. A practice that appears similar to a prohibition is not necessarily prohibited, and a practice that appears different can still fall within a prohibition where the conditions are met. The Article 5 screen must be done on the facts, with the Article text in hand.

What does the Act permit, subject to duties?

The Act permits AI systems that are not prohibited, subject to the duties that apply to their category. High risk systems can be placed on the market and used, subject to the high risk requirements, conformity assessment, registration and post market duties. Transparency systems can be used, subject to the Article 50 duties to inform, mark and disclose. General purpose models can be placed on the market, subject to the Chapter V duties for documentation, copyright, training summary and, for systemic risk models, evaluation and incident reporting. Non high risk systems can be used, subject to the prohibition and transparency screens and to other law. Innovation can be tested in sandboxes and in real world conditions, subject to the Chapter VI conditions. The Act is therefore a framework for placing AI on the market and using it, with duties that scale with risk, not a ban on the technology. A provider that treats the Act as a ban will avoid the market altogether, and will miss the opportunity to place a lawful system with the right file. A provider that treats the Act as permitting everything, subject only to a general duty to be careful, will miss the specific duties that apply to its category. The correct approach is to identify the category, identify the duties, and meet them with evidence.

For the broader regulatory framework, see questions and misconceptions overview.

How should a prohibition screen be done?

A prohibition screen should be done first, before classification and before conformity planning, because a prohibited practice cannot be made lawful by a conformity file. The screen should list each Article 5 prohibition, state the facts of the system that are relevant to it, and record the conclusion with reasons, including the conditions, exceptions and application dates considered. The screen should be done by a person who has read the Article 5 text, not by a person who has only read a summary, and should be reviewed by a second person where the practice is borderline. The screen should consider the intended purpose, the actual use, the data, the affected persons, the context, and the marketing and instructions, because a practice that is presented as one thing can operate as another. The screen should be updated when the system, the use or the law changes, and should be kept with the system inventory. Where the screen is borderline, or where the practice is close to a prohibition, the system should be paused and the position confirmed with a competent authority or legal adviser before the system is placed on the market or put into service. A prohibition screen that is done after launch, or that is recorded only as a single word, does not show that the prohibition was considered.

What about the new December 2026 prohibitions?

The two new prohibitions added by Regulation (EU) 2026/1744 apply from 2 December 2026, and are described in this guide only in general terms, as required by the verified date list. Article 5(1)(ba) concerns AI systems used to generate or manipulate non consensual intimate material, and Article 5(1)(bb) concerns material within Directive 2011/93/EU, child sexual abuse material, subject to the consent and defence conditions and the other provisions in the Article. This guide does not reproduce graphic detail, does not give examples involving identifiable people, and does not link to offending material. A system that generates or manipulates image, audio or video content should be screened against those prohibitions, with the consent, purpose and safeguards examined, and with legal advice sought where the system can generate intimate or sexual content. The screen should be done before the 2 December 2026 date for systems that will be on the market on that date, and should be recorded with the Article text and the facts. A general purpose model or a generative system that is not designed for intimate content can still raise the prohibition question where it can be used to generate such content, and the provider and deployer duties for preventing misuse, for transparency and for marking should be examined alongside the prohibition screen.

What should a reader do if a system might be prohibited?

If a system might be prohibited, the reader should pause the placing on the market or putting into service, record the facts and the Article 5 analysis, and confirm the position with a competent authority or legal adviser. It should not rely on a disclaimer, a terms clause, a user warning or a technical filter as a substitute for the legal analysis, although those measures can be relevant to the facts and to the safeguards. It should not proceed on the basis that the prohibition is unlikely to be enforced, or that other providers are doing the same thing, because enforcement practice does not change the legal position. It should consider whether the system can be redesigned to remove the prohibited practice, whether the use can be narrowed to a lawful purpose, or whether the system should not be placed on the market at all. The decision, the advice and the action should be recorded with dates. A system that is redesigned should be re screened, with the new facts and the new Article 5 analysis recorded. A system that is not redesigned and that remains borderline should not be launched on the basis of a general guide, including this one.

Continue with misconception: all AI is high risk.

What does this correction not decide?

This correction does not decide whether any named practice is prohibited, does not reproduce Article 5 in full, and does not describe the new December 2026 prohibitions beyond the general terms required by the verified date list. Prohibition decisions need the full Article 5 text, the facts of the practice, and confirmation from a competent authority or legal adviser where the practice is borderline. This guide corrects the general misconception that the Act bans AI, not the prohibition position for a particular practice.

Frequently asked questions

Continue through this AI Act cluster

• questions and misconceptions overview
• misconception: all AI is high risk
• worked example: customer chatbot
• screening order and evidence record
• prohibited practices in Article 5
• transparency duties