Skip to main content
Praxikon
Agent check
Article 5 of 1134%
Nederlands

Article 5: Prohibited AI practices

Article 5 prohibits AI practices that carry an unacceptable risk, such as harmful manipulation, social scoring, emotion recognition in the workplace and in education, and untargeted scraping of facial images. The prohibitions have applied since 2 February 2025, and an infringement can be fined since 2 August 2025 with up to 35 million euro or 7 percent of worldwide annual turnover. From 2 December 2026 two prohibitions are added: AI for non-consensual intimate imagery and for child sexual abuse material.

, AI and privacy legal adviserobligations checked against the source on
Check your own case for free

Praxikon tracks Article 5 (Prohibited AI practices) under the EU AI Act, citing the source for every statement.

EU Official:
ActiveIn force since 2 Feb 2025
Title II: Prohibited AI PracticesApplies from: 2 February 2025

Which AI practices are prohibited under Article 5?

The prohibition covers placing such AI on the market, putting it into service and using it. It therefore reaches both the provider that builds the system and the organisation that deploys it. These are the prohibitions in Article 5(1), in plain terms and with one example each.

  1. Article 5(1) (a)

    Harmful manipulation and deception

    AI that uses subliminal techniques or purposefully manipulative or deceptive techniques to materially distort people’s behaviour, so that they take a decision they would not otherwise have taken and that causes, or is reasonably likely to cause, significant harm.

    Example: An app that uses hidden cues to push users into ever higher stakes, causing significant financial harm.

    Ordinary personalisation or persuasive advertising is not caught as such: the test is material distortion with significant harm.

  2. Article 5(1) (b)

    Exploiting vulnerabilities

    AI that exploits vulnerabilities due to age, disability or a specific social or economic situation, materially distorting behaviour with significant harm as the (likely) result.

    Example: A chatbot that steers older people with early dementia towards expensive financial products they cannot oversee.

  3. Article 5(1) (ba) (from 2 December 2026)

    Non-consensual intimate imagery

    AI that generates or manipulates realistic images, video, audio or similar material of an identifiable person’s intimate parts, or of such a person engaged in sexually explicit activities, without that person’s explicit consent.

    Example: A so-called nudify app that turns an ordinary photo of an identifiable person into a nude image.

    For a provider it is prohibited where this is the intended purpose, or where the system foreseeably does it without reasonable safeguards; for a deployer where it uses the system for that purpose (Article 5(1a)).

  4. Article 5(1) (bb) (from 2 December 2026)

    Child sexual abuse material

    AI that generates or manipulates child sexual abuse material within the meaning of Directive 2011/93/EU.

    Example: An image generator without safeguards that produces such material on request.

    The same scoping in Article 5(1a) applies: intended purpose, or a foreseeable outcome without reasonable safeguards.

  5. Article 5(1) (c)

    Social scoring

    Evaluating or classifying people or groups over time based on social behaviour or personal characteristics, where the score leads to detrimental treatment in a context unrelated to where the data came from, or to treatment that is unjustified or disproportionate. The ban covers public authorities and private parties alike.

    Example: A landlord who turns tenants away based on a score drawn from their social media behaviour.

    A credit score based on relevant financial data is not social scoring, but it is high-risk (Annex III point 5(b)).

    What is social scoring?

  6. Article 5(1) (d)

    Predicting crime from profiling alone

    Assessing or predicting whether someone will commit a criminal offence, based solely on profiling or personality traits.

    Example: A system that flags residents as future offenders based only on their neighbourhood, age and a personality profile.

    Not prohibited: AI that supports a human assessment already based on objective and verifiable facts linked to a criminal activity.

  7. Article 5(1) (e)

    Untargeted scraping of facial images

    Creating or expanding facial recognition databases through untargeted scraping of facial images from the internet or CCTV footage.

    Example: Harvesting profile photos from social media at scale to build a face search engine.

  8. Article 5(1) (f)

    Emotion recognition in the workplace and in education

    Inferring people’s emotions in the workplace and in education institutions, except where the system is intended for medical or safety reasons.

    Example: Software that infers from the voice or face of call-centre staff whether they are frustrated.

    Fatigue detection that does not infer emotions falls outside the definition of emotion recognition (Article 3(39), recital 18).

    Emotion recognition in the workplace: what is and is not allowed

  9. Article 5(1) (g)

    Biometric categorisation by sensitive traits

    Categorising people on the basis of biometric data to deduce or infer their race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation.

    Example: Camera software that infers visitors’ religion from facial images.

    Not prohibited: labelling or filtering lawfully acquired biometric datasets, and categorisation in the area of law enforcement.

  10. Article 5(1) (h)

    Live facial recognition in public by the police

    Real-time remote biometric identification in publicly accessible spaces for law enforcement.

    Example: Police running live facial recognition on a busy square and matching everyone against a database.

    Only allowed where strictly necessary for three objectives: a targeted search for victims and missing persons, preventing a specific threat to life or a terrorist attack, or locating suspects of serious offences listed in Annex II. Prior authorisation by a judicial or independent authority is required; in a duly justified urgency, use may start if authorisation is requested within 24 hours (Article 5(3)).

Since when do the prohibitions apply?

The prohibitions have applied since 2 February 2025 (Article 113(a)). An infringement can be fined since 2 August 2025, when the penalty provisions became applicable (Article 113(b)). The Digital Omnibus (Regulation (EU) 2026/1744) did not postpone the existing prohibitions.

The two new prohibitions in points (ba) and (bb), on non-consensual intimate imagery and child sexual abuse material, were added by that Omnibus and apply from 2 December 2026 (Article 113(a)).

How high is the fine?

Up to 35 million euro or 7 percent of worldwide annual turnover, whichever is higher (Article 99(3)). That is the highest fine ceiling in the AI Act.

For SMEs, start-ups included, the lower of the two amounts applies (Article 99(6)). For small mid-cap enterprises that reduction does not apply to a prohibited practice: their rule (Article 99(6a)) only covers fines under paragraphs 4 and 5.

Check it yourself or have it checked

Sources

Official text

||

Source: EUR-Lex, Regulation (EU) 2024/1689. Text reproduced verbatim.

Download AI Act (PDF) →

Decision tree

Does your application fall under a prohibited practice?

Check per concrete use and context, not per technique: the same technique can be prohibited in one place and permitted in another. The prohibitions apply to providers and deployers alike.

Question 1 · Art. 5(1)(a)

Does the system deploy subliminal techniques or purposefully manipulative or deceptive techniques that materially distort people’s behaviour by impairing their ability to make an informed decision, in a manner that causes or is reasonably likely to cause significant harm?

The objective or the effect counts. Ordinary persuasive advertising does not fall under this; hidden influence that leads to harmful decisions does.

Full diagram
  1. 1

    Does the system deploy subliminal techniques or purposefully manipulative or deceptive techniques that materially distort people’s behaviour by impairing their ability to make an informed decision, in a manner that causes or is reasonably likely to cause significant harm?

    Art. 5(1)(a)

    • Yes → Outcome A
    • No → Question 2
  2. 2

    Does the system exploit vulnerabilities of persons due to their age, disability or a specific social or economic situation, with the objective or effect of materially distorting their behaviour in a manner that causes or is reasonably likely to cause significant harm?

    Art. 5(1)(b)

    • Yes → Outcome B
    • No → Question 3
  3. 3

    Does the system generate or manipulate realistic images, video, audio or similar material of the intimate parts of an identifiable person, or of an identifiable person engaged in sexually explicit activities, without that person’s freely given, specific, informed, unambiguous and explicit consent, or child sexual abuse material or performance, and do you meet the test for your role?

    Art. 5(1)(b) bis en b ter

    • Yes, under the test for my role → Outcome C
    • No → Question 4
  4. 4

    Does the system evaluate or classify persons over a certain period of time based on their social behaviour or known, inferred or predicted personal characteristics, with the score leading to detrimental treatment in a context unrelated to where the data came from, or to detrimental treatment that is unjustified or disproportionate to their social behaviour or its gravity?

    Art. 5(1)(c)

    • Yes → Outcome D
    • No → Question 5
  5. 5

    Does the system assess the risk of a person committing a criminal offence, based solely on profiling or on personality traits and characteristics?

    Art. 5(1)(d)

    • Yes → Question 6
    • No → Question 7
  6. 6

    Does the system only support a human assessment of a person’s involvement in a criminal activity that is already based on objective and verifiable facts directly linked to that activity?

    Art. 5(1)(d)

    • Yes → Outcome F
    • No → Outcome E
  7. 7

    Does the system create or expand a facial recognition database through the untargeted scraping of facial images from the internet or CCTV footage?

    Art. 5(1)(e)

    • Yes → Outcome G
    • No → Question 8
  8. 8

    Does the system infer emotions or intentions of persons on the basis of their biometric data, such as face, voice or posture, in the workplace or in an education institution?

    Art. 5(1)(f) · Art. 3(39)

    • Yes → Question 9
    • No → Question 10
  9. 9

    Is the system intended to be put in place or placed on the market for medical or safety reasons?

    Art. 5(1)(f)

    • Yes → Outcome I
    • No → Outcome H
  10. 10

    Does the system individually categorise persons based on their biometric data to deduce or infer their race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation?

    Art. 5(1)(g)

    • Yes → Outcome J
    • No → Question 11
  11. 11

    Do you use real-time remote biometric identification systems in publicly accessible spaces for the purposes of law enforcement?

    Art. 5(1)(h)

    • Yes → Outcome K
    • No → Outcome L

Outcome A · Prohibited

Prohibited: harmful manipulation or deception

Placing this system on the market, putting it into service and using it has been prohibited since 2 February 2025. Infringing Article 5 carries the highest fine ceiling: 35 million euro or 7 percent of worldwide annual turnover, for SMEs the lower of the two.

Art. 5(1)(a) · Art. 99(3) · Art. 99(6)

Outcome B · Prohibited

Prohibited: exploitation of vulnerabilities

A system that exploits vulnerabilities due to age, disability or social or economic situation and thereby causes or is reasonably likely to cause significant harm has been prohibited since 2 February 2025.

Art. 5(1)(b) · Art. 99(3)

Outcome C · Prohibited

Prohibited from 2 December 2026: non-consensual intimate material and child sexual abuse material

For the provider the prohibition applies where this generation is the intended purpose, or where it is a reasonably foreseeable and reproducible outcome without significant technical modification and the system lacks reasonable and adequate safeguards to reliably prevent it and to correct reported misuse. For the deployer the prohibition applies where it uses the system for that purpose. The prohibition on child sexual abuse material does not apply where national law provides a “without right” defence. Apart from the AI Act, creating or distributing this material may already be a criminal offence.

Art. 5(1)(b) bis · Art. 5(1)(b) ter · Art. 5(1) bis · Art. 5(1) ter · art. 113 onder a

Outcome D · Prohibited

Prohibited: social scoring

Social scoring that leads to detrimental treatment in an unrelated context, or to unjustified or disproportionate treatment, has been prohibited since 2 February 2025. The prohibition applies to public and private parties alike.

Art. 5(1)(c)

Outcome E · Prohibited

Prohibited: predicting crime from a profile

Risk assessments that predict whether a person will commit a criminal offence, based solely on profiling or personality traits, have been prohibited since 2 February 2025.

Art. 5(1)(d)

Outcome F · Check further

Not prohibited; high risk in law enforcement

The prohibition does not apply to systems that support a human assessment already based on objective and verifiable facts. If the system is used by or on behalf of a law enforcement authority to assess the risk of offending or re-offending, it is listed in Annex III point 6(d) and is in principle high risk. If a private party uses it outside law enforcement, check the classification separately under Article 6.

Art. 5(1)(d) · Art. 6(2) · bijlage III punt 6 onder d

Outcome G · Prohibited

Prohibited: untargeted scraping for facial recognition

Creating or expanding facial recognition databases through untargeted scraping of facial images from the internet or CCTV footage has been prohibited since 2 February 2025, without exception.

Art. 5(1)(e)

Outcome H · Prohibited

Prohibited: emotion recognition at work or school

Emotion recognition in the workplace and in education has been prohibited since 2 February 2025, except for medical or safety reasons.

Art. 5(1)(f) · Art. 3(39)

Outcome I · Check further

Exception for medical or safety reasons, but high risk

With a genuine medical or safety reason the system does not fall under the prohibition. Emotion recognition is listed in Annex III point 1(c) and is then in principle high risk. Deployers must also inform the persons exposed (Article 50(3)).

Art. 5(1)(f) · bijlage III punt 1 onder c · Art. 50(3)

Outcome J · Prohibited

Prohibited: biometric categorisation on sensitive traits

This has been prohibited since 2 February 2025. The prohibition does not cover labelling or filtering of lawfully acquired biometric datasets, such as images, nor categorising biometric data in the area of law enforcement.

Art. 5(1)(g)

Outcome K · Prohibited

Prohibited, except in three strictly defined cases

Only where strictly necessary for the targeted search for victims or missing persons, the prevention of a specific and imminent threat or terrorist attack, or locating suspects of serious criminal offences listed in Annex II. Then a national legal basis, prior authorisation by a judicial or independent authority, a FRIA and registration are required.

Art. 5(1)(h) · Art. 5(2) · Art. 5(3) · Art. 5(5)

Outcome L · Does not apply

No prohibited practice found

Based on your answers, the application does not fall under Article 5. Record the check with the reasoning per point, then check whether the system is high risk (Article 6) and whether transparency obligations apply (Article 50).

art. 5 · art. 6 · art. 50

A decision tree simplifies. If you are unsure at a question, read the provision or have it assessed. Checked against the legal text on 1 October 2026.

Official guidance on this article

5

What does this mean for you?

Provider+
As a provider you may not place on the market or put into service an AI system that falls under one of the prohibitions in Article 5(1), such as harmful manipulation, exploiting vulnerabilities, social scoring or emotion recognition for the workplace or education. These prohibitions have applied since 2 February 2025. If your system generates images, video or audio, also check the prohibitions in points (ba) and (bb) that apply from 2 December 2026: without reasonable safeguards against non-consensual intimate imagery and child sexual abuse material the system can be caught (Article 5(1a)).
Deployer+
As a deployer you may not use AI for a prohibited practice, even if you bought the system. Screen per concrete use and context, and pay particular attention to emotion recognition of your own staff or in education: it is prohibited except for medical or safety reasons (Article 5(1)(f)). Record the outcome per system.
SME / Startup+
The prohibitions apply in full to small businesses too. The difference is the fine: for SMEs and start-ups the lower of 35 million euro or 7 percent of turnover applies (Article 99(6)). Most SMEs do not touch a prohibition, but check AI used for staff assessment, customer profiling and biometrics.
Public Sector+
Public authorities may not apply social scoring or predict crime based solely on profiling. Real-time remote biometric identification in publicly accessible spaces for law enforcement is prohibited, except for the three objectives in Article 5(1)(h), with prior authorisation by a judicial or independent authority (Article 5(3)).

Related recitals

Recital 29Primary

AI-enabled manipulative techniques can be used to persuade persons to engage in unwanted behaviours, or to deceive them by nudging them into decisions in a way that subverts and impairs their autonomy…

Recital 30Primary

Biometric categorisation systems that are based on natural persons’ biometric data, such as an individual person’s face or fingerprint, to deduce or infer an individuals’ political opinions, trade uni…

Recital 31Primary

AI systems providing social scoring of natural persons by public or private actors may lead to discriminatory outcomes and the exclusion of certain groups. They may violate the right to dignity and no…

Recital 32Primary

The use of AI systems for ‘real-time’ remote biometric identification of natural persons in publicly accessible spaces for the purpose of law enforcement is particularly intrusive to the rights and fr…

3334353637383940414243444546

Related tools

Penalty CalculatorAI Decision Tree

Related enforcement

  • •
    AI Act prohibited practices enter into force — no enforcement actions yetEuropean AI Office · Feb 2025↗
  • •
    CNIL fines Amazon €32 million for AI employee monitoringCNIL · Dec 2023↗
  • •
    Dutch DPA fines Clearview AI €30.5 millionAutoriteit Persoonsgegevens · Sept 2023↗
  • •
    Clearview AI fined €20 million by Italian DPAGarante per la protezione dei dati personali · May 2022↗
  • •
    Greek DPA fines Clearview AI €20 millionHellenic Data Protection Authority · May 2022↗

Related blog posts

Cross-references

Articles

Art. 96Art. 99

Recitals

2829303132333435

Annexes

Annex 2

Frequently asked questions

Which AI practices are prohibited under the AI Act?+
Article 5(1) prohibits harmful manipulation and deception (point a), exploiting vulnerabilities due to age, disability or a social or economic situation (point b), social scoring (point c), predicting crime based solely on profiling (point d), untargeted scraping of facial images for facial recognition databases (point e), emotion recognition in the workplace and in education, except for medical or safety reasons (point f), biometric categorisation by sensitive traits (point g) and real-time remote biometric identification in publicly accessible spaces for law enforcement, with narrow exceptions (point h). From 2 December 2026, AI for non-consensual intimate imagery (point ba) and for child sexual abuse material (point bb) are added.
What are the penalties for prohibited AI practices?+
An infringement of Article 5 can be fined up to 35 million euro or 7 percent of worldwide annual turnover, whichever is higher (Article 99(3)). For SMEs, start-ups included, the lower of the two applies (Article 99(6)). For small mid-cap enterprises that reduction does not apply to Article 5: Article 99(6a) only covers fines under paragraphs 4 and 5.
Since when do the prohibited AI practices apply?+
The prohibitions in Article 5 have applied since 2 February 2025 (Article 113(a)). Fines can be imposed since 2 August 2025, when the penalty provisions became applicable (Article 113(b)). The two prohibitions added by the Digital Omnibus (Regulation (EU) 2026/1744), on non-consensual intimate imagery and child sexual abuse material, apply from 2 December 2026.
Did the Digital Omnibus postpone Article 5?+
No. The Digital Omnibus (Regulation (EU) 2026/1744) did not postpone the existing prohibitions; they have applied since 2 February 2025. What was postponed are the high-risk obligations: to 2 December 2027 for the Annex III route and to 2 August 2028 for the Annex I route. The Omnibus did add two prohibitions that apply from 2 December 2026.
Can I still use emotion recognition in my company?+
Not on your own staff or in education: emotion recognition in the workplace and in education institutions is prohibited, except for medical or safety reasons (Article 5(1)(f)). In other situations, for example with customers, emotion recognition is not prohibited but it is high-risk (Annex III point 1(c)), and you must inform the people exposed to it (Article 50(3)). Fatigue detection that does not infer emotions falls outside the definition (Article 3(39), recital 18).
Is social scoring by companies also prohibited or only by governments?+
Both. Article 5(1)(c) prohibits AI that evaluates people or groups over time based on social behaviour or personal characteristics, where the score leads to detrimental treatment in a context unrelated to where the data came from, or to treatment that is unjustified or disproportionate. That applies to public authorities and private organisations alike.
What is the difference between social scoring and an ordinary credit assessment?+
A credit assessment based on relevant financial data, such as income and payment history, is not social scoring. Such a system is high-risk (Annex III point 5(b)). It becomes an Article 5 problem when behaviour from another context, such as social media, leads to detrimental treatment, or when that treatment is disproportionate.
Does an ordinary AI chatbot fall under the prohibited practices?+
Usually not. A chatbot is normally covered by the disclosure duty in Article 50(1): people must know they are talking to AI. A chatbot only falls under Article 5 if it harmfully manipulates people or exploits vulnerabilities (points a or b), or if it infers the emotions of staff or students (point f).
May the police use live facial recognition in public spaces?+
In principle not (Article 5(1)(h)). Only where strictly necessary for a targeted search for victims or missing persons, preventing a specific threat to life or a terrorist attack, or locating suspects of serious offences listed in Annex II. Prior authorisation by a judicial or independent authority is required (Article 5(3)).
How do I know if my AI system falls under the prohibited practices?+
Screen per concrete use and context, not per technology: the same technique can be prohibited in the workplace and permitted elsewhere. Walk through the prohibitions in Article 5(1) for each system and record the outcome, also when it is that there is no prohibited practice. The Praxikon case check does that first screening for free, and the European Commission's guidelines of 4 February 2025 give examples per prohibition.
What is the difference between prohibited AI and high-risk AI?+
Prohibited AI (Article 5) may not be used in the EU at all: there is no way to comply. High-risk AI (Article 6 and Annex III) may be used, but only with requirements on risk management, data, documentation, transparency and human oversight. For the Annex III route those requirements apply from 2 December 2027.

What Article 5 requires in practice

Connections

What connects to Article 5 AI Act

Themes where this returns

Where this meets the other law

The counterpart in the other law

GDPR interpretation that also applies here6 of 10