Prohibited AI practices are the uses of artificial intelligence that Regulation (EU) 2024/1689, the AI Act, bans across the Union because it considers them an unacceptable risk. They are set out in Article 5 and have applied since 2 February 2025. There are eight, among them subliminal manipulation, social scoring, emotion recognition in the workplace and building facial recognition databases from images scraped from the internet. The Digital Omnibus on AI, in force since 27 July 2026, adds two more from 2 December 2026. They concern the generation of non-consensual intimate images and of child sexual abuse material. The ban reaches whoever places the system on the market and also the company that uses it. And it is the most expensive infringement in the regulation, with fines of up to 35 million euros or 7 % of worldwide turnover.
If you need an AI lawyer in Spain for your business, request a free initial assessment.
Article 5(1) of Regulation (EU) 2024/1689 prohibits placing on the market, putting into service or using AI systems for eight purposes. Each letter has its own conditions, and in several of them the ban only applies where there is significant harm or a specific purpose.
| Letter | What is banned | Condition that triggers it |
|---|---|---|
| a) | Subliminal, manipulative or deceptive techniques | That they materially distort behaviour and cause, or are reasonably likely to cause, significant harm |
| b) | Exploiting vulnerabilities due to age, disability or a social or economic situation | The same distortion and harm |
| c) | Social scoring of people based on their behaviour or characteristics | That it leads to detrimental treatment in other contexts or treatment that is disproportionate |
| d) | Predicting the risk that a person will commit a crime | That it is based solely on their profile or personality traits |
| e) | Creating or expanding facial recognition databases | Through untargeted scraping of images from the internet or CCTV footage |
| f) | Inferring a person’s emotions | In the workplace or in education institutions, except for medical or safety reasons |
| g) | Biometric categorisation to deduce sensitive data | Race, political opinions, trade union membership, religion, sex life or sexual orientation |
| h) | Real-time remote biometric identification | In publicly accessible spaces and for law enforcement, save for listed exceptions |
In February 2025 the Commission published guidelines on these practices, with examples of what falls inside and outside each one. They are not binding, but they are the reference the authorities will use to interpret each letter.
Regulation (EU) 2026/1744, the Digital Omnibus on AI, adds two letters to Article 5(1) that apply from 2 December 2026. Letter (ba) bans systems that generate or manipulate realistic images, videos or audio of the intimate parts of an identifiable person without their explicit consent. It also bans those that show the person engaged in sexually explicit activities without that consent. Letter (bb) bans systems that generate child sexual abuse material within the meaning of Directive 2011/93/EU.
The new paragraph 1a defines who the ban reaches. For the provider, placing the system on the market is only prohibited in two cases.
For whoever uses the system, the ban applies when it is used for the purpose of generating that material. Paragraph 1b clarifies that there is no manipulation if the change does not increase the exposure of any intimate part or alter the nature of the activity shown.
In practice this mainly affects those who develop or integrate image, video or voice generators. Safeguards against foreseeable misuse stop being good practice and start to decide whether the product is lawful in the Union.
Much of Article 5 is aimed at public authorities, such as real-time biometric identification or crime prediction. For a private company, four letters account for almost all the risk.
Letter f bans using AI to infer people’s emotions in the workplace or in education institutions. The only exception is a system installed for medical or safety reasons. It affects voice analytics tools in call centres, facial analysis in meetings and tools that measure staff mood. Outside work and education, emotion recognition is not banned, but it is high-risk under Annex III.
Letters a and b target designs that push a person into a decision they would not otherwise have taken and that harms them. A recommender system or a sales assistant is not banned for being persuasive. It is banned if it uses subliminal or deceptive techniques, or exploits the user’s age or economic situation, and causes significant harm. The bar is high, but dark patterns aimed at minors or at people in financial difficulty are in the danger zone.
Letter c is not limited to governments. A company that scores its customers on behaviour unrelated to the service and denies them treatment as a result may fall within it. Letter e bans building facial databases with images collected en masse from the internet or from cameras. It affects facial recognition providers and anyone who buys their databases.
Breaching a prohibition in Article 5 carries the highest fine in the regulation. It can reach 35 million euros or 7 % of total worldwide annual turnover for the preceding financial year, whichever is higher. For SMEs and start-ups the lower of the two applies. The detail of each tier is in our analysis of AI Act penalties.
Spain does not yet have its own penalty regime. The Organic Law bill on the proper use and governance of artificial intelligence, published by the Spanish Congress on 12 June 2026, classed prohibited practices as a very serious infringement. It also shared their supervision among several authorities.
The bill lapsed when the Spanish Parliament was dissolved, as published in the Official State Gazette (BOE) on 6 October 2026, and the government formed after the elections of 29 November will have to reintroduce it. The ban, on the other hand, has applied directly since February 2025 and does not depend on that law.
This case is fictitious. Voxia Servicios, S.L. runs a call centre in Valencia with 90 agents. Its supplier offers a module that analyses each agent’s voice and gives them a stress and irritation score. The idea is to use it in monthly appraisals.
That module infers the emotions of workers in their workplace, so its use is banned by Article 5(1)(f). It makes no difference that the score is only indicative, or that the supplier sells it as a wellbeing tool, because it does not serve a medical or safety purpose. Voxia could analyse customers’ voices to detect urgent calls, because customers are not in their workplace. That second use, however, would be high-risk under Annex III and would carry its own obligations.
If Voxia used the module on its staff, it would also trigger the duty of algorithmic information under Spain’s Royal Decree 723/2026. We explain it in our guide to algorithmic transparency for workers in Spain. The full framework of the regulation is in our AI Act compliance guide for businesses.
The eight practices in Article 5(1) of the AI Act have been banned since 2 February 2025. The two added by the Digital Omnibus on AI apply from 2 December 2026. They are the generation of non-consensual intimate images and of child sexual abuse material. Fines for breaching them can reach 35 million euros or 7 % of worldwide turnover.
No, unless the system serves a medical or safety purpose. Article 5(1)(f) of the AI Act bans using AI to infer people’s emotions in the workplace and in education institutions. The ban applies even if the stated purpose is staff wellbeing or the score is only indicative.
Not for being persuasive. Article 5(1)(a) only bans systems that use subliminal, manipulative or deceptive techniques to materially distort a person’s decision. It also requires that they cause, or are reasonably likely to cause, significant harm. A transparent sales chatbot does not fall within it, although it must tell users they are talking to an AI under Article 50.
From 2 December 2026, the ban reaches the user who uses the system for the purpose of generating that content. It reaches the provider if that generation is the system’s purpose or a foreseeable result without reasonable safeguards. A company that only uses the tool for other purposes is not caught by the ban, but it is advisable to set internal rules of use.
Spain does not yet have its own penalty regime for the AI Act. The Organic Law bill on the governance of AI assigned supervision to AESIA as a general rule. It reserved facial databases and biometric categorisation to the Spanish Data Protection Agency, and law enforcement uses to the General Council of the Judiciary. The bill lapsed when the Spanish Parliament was dissolved in October 2026 and will have to be reintroduced.
The AI Act bans already apply, and the Omnibus extends the list in December 2026. For most businesses the risk does not lie in deliberate use. It lies in a module that comes built into an HR, customer service or marketing tool. Checking what each system does before switching it on is cheaper than withdrawing it later. At Innovatech we review how your systems fit Article 5 as part of our AI legal advisory service. Write to us and we will give you a free initial assessment.
Marta Suárez-Mansilla is Managing Partner of Innovatech Legal and a Spanish lawyer (abogada), Madrid Bar (ICAM), working in technology law. She completed Harvard Law School's Copyright course and BerkeleyX's Blockchain programme, and has advised technology companies for more than eight years.