EU AI Act: Prohibited AI Practices — The Ten Things You May Not Do With AI in Europe

What Article 5 Prohibits — The Unacceptable Risk Tier

The EU AI Act regulates AI through a risk pyramid. Most systems face no obligations; some face transparency duties (Article 50); high-risk systems face strict controls (Chapter III). At the top sits a category the law refuses to permit at all: unacceptable risk.

Article 5 lists AI practices considered incompatible with fundamental rights and Union values — human dignity, freedom, equality, non-discrimination, and privacy. These practices may not be placed on the market, put into service, or used in the EU. There is no conformity assessment pathway, no registration option, and no grandfathering for systems deployed before the ban.

Three features make Article 5 the sharpest instrument in the AI Act. First, it took effect earlier than any other substantive obligation — on February 2, 2025, six months after entry into force. Second, it carries the highest penalty tier: €35 million or 7% of worldwide turnover under Article 99(3). Third, it binds both providers and deployers — the company that builds the system and the company that uses it.

The prohibitions also reach beyond Europe. Under Article 2, a provider established anywhere in the world is covered if its system is placed on the EU market, and even a non-EU deployer is covered where the output of the system is used in the EU.

The Eight Original Prohibitions at a Glance

Article 5(1) sets out eight prohibited practices, lettered (a) through (h). Each has precise conditions — and several have narrow, explicitly drafted exceptions.

Manipulation and Exploitation — Art. 5(1)(a) and (b)

Subliminal and Manipulative Techniques — (a)

The first prohibition targets AI systems that deploy subliminal techniques beyond a person’s consciousness, or purposefully manipulative or deceptive techniques. Three cumulative conditions must all be met:

• The system uses subliminal, purposefully manipulative, or deceptive techniques
• The technique materially distorts behaviour, appreciably impairing the ability to make an informed decision
• The distorted behaviour causes, or is reasonably likely to cause, significant harm

The Commission guidelines draw the line between unlawful manipulation and lawful persuasion: advertising, recommendation, and personalisation that operate transparently remain permitted. The prohibition bites when techniques work covertly — below the threshold of awareness — and produce significant physical, psychological, or financial harm.

Exploiting Vulnerabilities — (b)

The second prohibition covers AI systems that exploit vulnerabilities of a person or group due to their age, disability, or specific social or economic situation, with the objective or effect of materially distorting behaviour in a manner that causes or is reasonably likely to cause significant harm.

Practical examples from the guidelines include AI toys that nudge children toward dangerous behaviour, and predatory lending or pricing systems that target people in financial distress. Unlike (a), this prohibition has no persuasion carve-out — targeting a protected vulnerability with harmful behavioural distortion is banned outright.

Social Scoring and Crime Prediction — Art. 5(1)(c) and (d)

Social Scoring — (c)

AI systems that evaluate or classify people over time based on their social behaviour or known, inferred, or predicted personal characteristics are prohibited where the resulting score leads to either:

• Detrimental treatment in social contexts unrelated to the context in which the data was originally generated or collected, or
• Detrimental treatment that is unjustified or disproportionate to the social behaviour or its gravity

Critically, the ban applies to private companies as well as public authorities. It does not automatically prohibit credit scoring, insurance pricing, or fraud detection based on relevant data in the original context — the test is cross-context misuse or disproportionate treatment based on social data.

Individual Crime Prediction — (d)

AI systems that assess or predict the risk of a natural person committing a criminal offence, based solely on profiling or on assessing personality traits and characteristics, are prohibited.

The provision contains an explicit carve-out: AI systems used to support the human assessment of involvement in criminal activity remain permitted where that assessment is already based on objective and verifiable facts directly linked to a criminal activity. Place-based predictive policing — forecasting where offences may occur rather than who will commit them — also falls outside the prohibition, though it may be high-risk under Annex III.

The Biometric Prohibitions — Art. 5(1)(e), (f), (g)

Real-Time Remote Biometric Identification — Art. 5(1)(h)

The most heavily negotiated provision of the entire AI Act. The use of real-time remote biometric identification (RBI) systems — live facial recognition — in publicly accessible spaces for law enforcement purposes is prohibited, unless one of three exhaustively listed situations applies:

Safeguards That Apply Even Within the Exceptions

• Art. 5(2) — Use only to confirm the identity of the specifically targeted individual; a fundamental rights impact assessment (Article 27) and registration in the EU database (Article 49) are required
• Art. 5(3) — Prior authorisation by a judicial authority or independent administrative authority; in duly justified emergencies, authorisation may follow within 24 hours or use must stop
• Art. 5(4) — Each use must be notified to the market surveillance authority and the data protection authority
• Art. 5(5) — A Member State may only allow RBI if it has enacted national law; it must notify the Commission within 30 days, and may choose stricter rules — including not allowing RBI at all

New for 2026: Non-Consensual Intimate Material and CSAM

The Digital Omnibus on AI — adopted by the European Parliament on June 16, 2026, approved by the Council on June 29, 2026, and signed on July 8, 2026 — adds a new prohibition to Article 5, the first expansion of the banned list since the AI Act was adopted.

What Is Now Prohibited

AI systems that generate or manipulate non-consensual intimate images, video, audio, or similar material, or child sexual abuse material (CSAM).

The Provider-Side Standard

For providers, the prohibition applies where such generation is a reasonably foreseeable and reproducible outcome of the system that cannot be avoided without substantial technical modification. In practice, this requires generative AI providers to implement and maintain technical safeguards — content filters, safety classifiers, and abuse monitoring — that prevent these outputs from being a reproducible result of ordinary use.

Transition Period

The addition responds to the rapid growth of AI-generated intimate-image abuse and complements existing EU instruments on child sexual abuse and gender-based violence. For deployers, generating such material with any AI system was already unlawful under criminal law across Member States — the Omnibus places responsibility on the systems themselves.

The Commission Guidelines on Prohibited Practices

Two days after the prohibitions took effect, the European Commission approved Guidelines on Prohibited AI Practices (C(2025) 884, February 4, 2025). The formally adopted all-language version followed on July 29, 2025 (C(2025) 5052). At over 130 pages, the guidelines are the most detailed interpretive document on Article 5 available.

What the Guidelines Clarify

• Lawful persuasion vs. manipulation — transparent advertising, personalisation, and nudging remain permitted; covert techniques causing significant harm do not
• Scope of each prohibition — concrete examples and non-examples for all eight practices, including edge cases like AI companions, games, and wellness apps
• Interplay with other law — how Article 5 operates alongside the GDPR, the Digital Services Act, and consumer protection rules
• The medical/safety exception for emotion inference — interpreted narrowly, covering fatigue detection and clinical support, not HR analytics
• Out-of-scope uses — systems not placed on the EU market, and practices lacking the required harm element

Enforcement and Penalties: The €35M / 7% Tier

Article 99(3) reserves the highest fines in the AI Act for Article 5 violations. Penalties have been applicable since August 2, 2025, when Member States were required to designate market surveillance authorities and lay down their national penalty rules.

How Enforcement Works

• National market surveillance authorities police the prohibitions in each Member State — unlike GPAI model obligations, which the Commission’s AI Office enforces centrally
• Data protection authorities act in parallel where prohibited practices process personal data, since the GDPR applies cumulatively
• Member State penalty laws set the specific rules within the Article 99 maximums; several states also provide criminal sanctions for related conduct
• SME proportionality — for SMEs and startups, the lower of the fixed amount or percentage applies (Article 99(6))

Penalty Tiers Across the AI Act

Comparison: Prohibited vs High-Risk vs Transparency vs GPAI

The four obligation frameworks of the AI Act differ in who is bound, when, and how hard the penalties bite.

Read our companion guides: Article 50 Transparency Obligations and GPAI Model Obligations.

Five-Step Article 5 Screening Checklist

Because prohibited practices cannot be fixed with documentation or registration, the only compliance strategy is to make sure none of your systems performs one. Screen systematically:

Complete Article 5 Timeline

Frequently Asked Questions

Article 5 Readiness Checklist

Use this checklist to confirm that no system in your organisation crosses an Article 5 line.