RegulationObligations

Prohibited AI practices under the EU AI Act

The EU AI Act prohibits eight AI practices under Article 5. Review the list of banned unacceptable risk systems, the 2025 deadline, and maximum penalties.

, AI Researcher10 min read
In this article8
  1. The Article 5 threshold: unacceptable risk and the February 2025 deadline
  2. Manipulative techniques and exploitation of vulnerabilities
  3. Social scoring and predictive policing
  4. Facial recognition scraping and real-time remote biometrics
  5. Emotion inference and sensitive biometric categorisation
  6. Enforcement and the 35 million EUR penalty framework
  7. Frequently asked questions
  8. Sources
Passengers passing through a facial-recognition gate at a metro station
Real-time remote biometric identification in a publicly accessible space — the practice Article 5 restricts most tightly. Photo: Shwangtianyuan, CC BY-SA 4.0, via Wikimedia Commons

The Article 5 threshold: unacceptable risk and the February 2025 deadline

The European Union classifies AI systems into distinct risk tiers under Regulation (EU) 2024/1689 (the AI Act), reserving the category of unacceptable risk for technologies that contravene fundamental rights. The Act entered into force on 1 August 2024, and the prohibited AI practices set out in Chapter II, Article 5 entered into application on 2 February 2025[1][2]. That date fell well before the Act's general date of application on 2 August 2026, establishing an immediate, binding compliance threshold across the European single market[2].

The legal prohibition applies comprehensively to placing on the market, putting into service, or using any system falling within the banned categories under Article 5(1). These restrictions bind providers, deployers, importers, and distributors, as structured in the AI Act provider definition. On 4 February 2025, the European Commission published official interpretative guidelines to support uniform enforcement and legal clarity across Member States[3].

Risk CategoryArticle ReferenceApplication DateCore Regulatory Mechanism
Unacceptable RiskArticle 52 February 2025Outright statutory prohibition on deployment, market placement, and use
General-Purpose AI (GPAI)Articles 51 to 562 August 2025Model technical documentation, copyright policies, and systemic risk evaluations
Transparency / Specific RiskArticle 502 August 2026 (general application of the Act)Labelling of AI-generated content and user notification obligations
High Risk (Annex III sensitive-area use cases)Articles 6 to 492 December 2027, as extended by the AI OmnibusEx-ante conformity assessments, risk management systems, and EU database registration
High Risk (systems embedded in regulated products, Annex I)Articles 6 to 492 August 2028Conformity assessment integrated into existing product safety regimes

Unlike high-risk AI architectures that permit deployment following conformity assessments and risk management procedures, unacceptable risk systems cannot be brought into compliance through technical safeguards, user consent, or administrative documentation. Any commercial distribution or operational use within EU jurisdiction constitutes an immediate statutory violation.

Manipulative techniques and exploitation of vulnerabilities

Article 5(1)(a) and Article 5(1)(b) prohibit AI systems designed to distort human decision-making and exploit individual or demographic vulnerabilities. Article 5(1)(a) targets systems that deploy subliminal techniques beyond a person's consciousness or purposefully manipulative or deceptive techniques. It bites where the technique has the objective or the effect of materially distorting behaviour, so that a person takes a decision they would not otherwise have taken.

Article 5(1)(b) addresses the exploitation of the vulnerabilities of a natural person or a specific group due to their age, disability, or a specific social or economic situation. This provision covers systems that target such characteristics to undermine rational agency, such as AI-driven commercial mechanisms targeting minors or automated interfaces designed to exploit personal debt distress.

  • Subliminal or purposefully manipulative techniques: The system uses techniques beyond a person's consciousness, or purposefully manipulative or deceptive techniques.
  • Vulnerability targeting: The system exploits vulnerabilities due to age, disability, or a specific social or economic situation.
  • Material behavioural distortion: The technique has the objective or effect of materially distorting behaviour by appreciably impairing the ability to make an informed decision.
  • Threshold of significant harm: The distorted behaviour causes, or is reasonably likely to cause, significant harm to that person, another person, or a group of persons.

The statutory standard requires demonstrating both material distortion and a reasonable likelihood of significant harm. Standard commercial recommendation engines, basic user interface personalisation, and conventional digital marketing that do not impair informed decision-making or generate significant harm remain outside the Article 5 prohibition, though they remain subject to EU consumer protection law and Digital Services Act rules.

Social scoring and predictive policing

Article 5(1)(c) establishes a comprehensive ban on AI-driven social scoring of natural persons. The statute prohibits systems deployed by public authorities or private entities that evaluate or classify individuals or groups over time based on their social behaviour or known, inferred, or predicted personality characteristics. The prohibition applies when the evaluation leads to detrimental or unfavourable treatment in social contexts unrelated to where the data was originally generated, or treatment that is unjustified or disproportionate to the behaviour's gravity.

Article 5(1)(d) bans AI systems that assess or predict the risk of a natural person committing a criminal offence where that assessment is based solely on profiling or on assessing their personality traits and characteristics. This prohibition prevents automated profiling based on personality traits, geographic location histories, or demographic indicators from serving as the basis for criminal suspicion or enforcement actions.

  • Cross-context social evaluation: social scoring that leads to detrimental or unfavourable treatment in social contexts unrelated to the contexts in which the data was originally generated or collected is prohibited[1].
  • Disproportionate penalties: Automated scoring models that impose civil, social, or economic disadvantages out of proportion to specific conduct are unlawful.
  • Individual criminal risk profiling: Algorithmic assessments predicting an individual's criminality risk based solely on profiling or personality traits are banned under Article 5(1)(d).
  • Fact-based human assessment exception: AI tools used strictly to support human assessments based on objective, verifiable facts directly linked to criminal activity remain permissible.

The line falls on data provenance and human oversight: tools that analyse objective evidence, such as transaction records in a fraud investigation, remain lawful, while systems assigning general criminal propensity scores from demographic attributes fall under the ban.

Facial recognition scraping and real-time remote biometrics

Under Article 5(1)(e), the EU AI Act strictly prohibits placing on the market, putting into service, or using AI systems that create or expand facial recognition databases through the untargeted scraping of facial images from the internet or CCTV footage.

Article 5(1)(h) prohibits the use of real-time remote biometric identification (RBI) systems in publicly accessible spaces for the purposes of law enforcement, unless and in so far as such use is strictly necessary for one of three exhaustively listed objectives[1]. Because real-time biometric tracking across public venues presents profound risks to privacy and assembly, deployment is banned by default across all Member States, subject only to those narrow statutory exceptions.

Objective CategoryStatutory Scope under Article 5(1)(h)Mandatory Authorisation Requirements
Search for VictimsTargeted search for specific victims of abduction, trafficking in human beings or sexual exploitation of human beings, as well as the search for missing personsPrior judicial or independent administrative authorisation; in urgent cases, authorisation requested at the latest within 24 hours
Threat PreventionPrevention of a specific, substantial and imminent threat to the life or physical safety of natural persons, or a genuine and present or genuine and foreseeable threat of a terrorist attackFundamental rights impact assessment; registration in the EU database; temporal, geographic and personal limits
Suspect IdentificationLocalisation or identification of a person suspected of an offence referred to in Annex II, punishable in the Member State concerned by a custodial sentence or detention order for a maximum period of at least four yearsPrior authorisation based on objective evidence; no adverse legal decision based solely on the system's output

Where an exception applies, each use must be subject to prior authorisation granted by a judicial authority or an independent administrative authority whose decision is binding, and the law enforcement authority must have completed a fundamental rights impact assessment and registered the system in the EU database. Every use must also be notified to the relevant market surveillance authority and the national data protection authority, which submit annual reports on such use to the Commission[1].

Emotion inference and sensitive biometric categorisation

Article 5(1)(f) bans the placing on the market, putting into service, or use of AI systems intended to infer emotions of natural persons in workplace and educational environments. This prohibition addresses software designed to evaluate worker productivity, stress, attention, or student engagement through facial expression analysis, voice tone tracking, or physiological monitoring.

Article 5(1)(g) prohibits biometric categorisation systems that analyse biometric data to deduce or infer sensitive individual characteristics. Prohibited categories include race, political opinions, trade union membership, religious or philosophical beliefs, sex life, or sexual orientation. These prohibitions prevent automated profiling that exposes individuals to systemic discrimination across enterprise AI infrastructure environments.

  • Workplace and education ban: Deploying emotion recognition software on employees, job applicants, or students is prohibited under Article 5(1)(f).
  • Medical and safety carve-out: Emotion detection systems deployed strictly for medical purposes or operational safety (such as driver fatigue monitoring) remain lawful.
  • Protected attribute categorisation ban: Inferring sensitive characteristics including political beliefs, religion, race, and sexual orientation is prohibited under Article 5(1)(g).
  • Law enforcement and data labelling exceptions: Categorisation of biometric data by law enforcement and the technical labelling or filtering of lawfully acquired biometric datasets remain permitted.

These restrictions apply across corporate monitoring tools, hiring platforms, and classroom surveillance software. Organisations integrating computer vision or vocal analytics into human resource workflows must eliminate emotion-tracking modules to avoid regulatory exposure.

Enforcement and the 35 million EUR penalty framework

Infringements of the Article 5 prohibitions carry the highest tier of administrative sanctions under the EU AI Act.[5] Under Article 99(3), non-compliance with the prohibition of the AI practices referred to in Article 5 is subject to administrative fines of up to EUR 35 000 000 or, if the offender is an undertaking, up to 7% of its total worldwide annual turnover for the preceding financial year, whichever is higher[4]. For SMEs, including start-ups, Article 99(6) provides that each fine shall be up to the percentages or amount referred to in paragraphs 3, 4 and 5, whichever thereof is lower[4].

Enforcement runs through the national market surveillance authorities that every Member State must establish or designate under Article 70.[7] At Union level, the European AI Office, established within the European Commission, supports implementation and enforces the rules on general-purpose AI models[6]. Market surveillance bodies hold the investigatory powers of Regulation (EU) 2019/1020, including the authority to demand technical documentation and to order that a non-compliant system be withdrawn or recalled from the market.

Infringement CategoryArticle ReferenceStandard Enterprise Penalty CapSME and Start-up Provision
Prohibited AI practices (top tier, the only tier applying to Article 5)Article 5 / Article 99(3)Up to EUR 35 000 000, or 7% of total worldwide annual turnover for the preceding financial year, whichever is higherWhichever of the two is lower, under Article 99(6)
High-risk and other operator obligations (separate, lower tier, not applicable to Article 5 breaches)Articles 16, 22-24, 26, 31, 33, 34, 50 / Article 99(4)Up to EUR 15 000 000, or 3% of turnover, whichever is higherWhichever of the two is lower, under Article 99(6)
Incorrect, incomplete or misleading information to authorities (separate, lowest tier, not applicable to Article 5 breaches)Article 99(5)Up to EUR 7 500 000, or 1% of turnover, whichever is higherWhichever of the two is lower, under Article 99(6)

When determining fine amounts, authorities must assess aggravating and mitigating circumstances under Article 99(7), including the gravity and duration of the infringement, prior violations, financial benefits gained, and the degree of cooperation with regulators[4]. Given the extraterritorial scope of the AI Act, international providers deploying systems accessible within the single market face direct financial exposure under this enforcement structure.

Frequently asked questions

What are the prohibited AI practices under the EU AI Act?

Article 5 of the EU AI Act bans eight AI practices: subliminal manipulation, exploiting vulnerabilities, social scoring, individual criminal risk assessment based solely on profiling, untargeted facial scraping, workplace and education emotion inference, sensitive biometric categorisation, and certain real-time remote biometric identification. The AI Omnibus (Regulation (EU) 2026/1744) added a further ban on AI systems that generate or manipulate non-consensual sexually explicit or intimate content and child sexual abuse material, which applies from 2 December 2026.

When do the EU AI Act prohibitions take effect?

The prohibitions on unacceptable risk AI systems under Article 5 entered into application on 2 February 2025, six months after the AI Act entered into force.

What is the penalty for using a prohibited AI system?

Under Article 99(3), non-compliance with the Article 5 prohibitions is subject to administrative fines of up to EUR 35 000 000 or, for an undertaking, up to 7% of total worldwide annual turnover for the preceding financial year, whichever is higher.

Are there exceptions to the ban on real-time biometric identification?

Yes. While generally prohibited for law enforcement in publicly accessible spaces, exceptions exist for strictly necessary uses: the targeted search for victims of abduction, trafficking or sexual exploitation and for missing persons, the prevention of specific imminent threats to life or a terrorist attack, and the localisation of suspects of serious offences listed in Annex II.

Can companies use AI to infer employee emotions?

No. Article 5(1)(f) prohibits AI systems used to infer the emotions of individuals in workplaces and education institutions, except where the system is intended to be put in place or on the market for medical or safety reasons.

Does the AI Act ban facial recognition databases?

Article 5(1)(e) prohibits creating or expanding facial recognition databases through the untargeted scraping of facial images from the internet or CCTV footage.

Sources

  1. Article 5: Prohibited AI practices ai-act-service-desk.ec.europa.eu
  2. AI Act digital-strategy.ec.europa.eu
  3. Commission publishes the Guidelines on prohibited artificial intelligence (AI) practices, as defined by the… digital-strategy.ec.europa.eu
  4. Article 99: Penalties ai-act-service-desk.ec.europa.eu
  5. The enforcement framework of the AI Act digital-strategy.ec.europa.eu
  6. European AI Office digital-strategy.ec.europa.eu
  7. Article 70: Designation of national competent authorities and a single point of contact ai-act-service-desk.ec.europa.eu

Rankings and scores on Optinest cannot be bought — no commercial relationship changes how a provider is assessed. How we score

Related reading

All Regulation →

No, Europe has not banned AI

Discover which AI practices are banned under the EU AI Act and how compliant EU-hosted infrastructure can unblock your enterprise generative AI pilots.

A police camera surveillance sign on a street in The Hague