Prohibited AI Practices under Article 5 of the AI Act: Every Ban Explained, Including the 2026 Additions

Prohibited AI practices under the EU AI Act are the ten uses of AI that Article 5(1) bans outright: eight since 2 February 2025 and two more from 2 December 2026. The eight older bans are points (a), (b) and (c) to (h); the two additions, points (ba) and (bb), target non-consensual intimate material and child sexual abuse material. No risk assessment, documentation or registration makes a listed practice lawful.

Article 5 opens the first risk tier of the EU AI Act, Regulation (EU) 2024/1689, in force since 1 August 2024 and amended by Regulation (EU) 2026/1744 of 8 July 2026, in force since 27 July 2026.

Compiled by the AIActGuide.net Editorial Team from the Official Journal text; how we check every statement: methodology.

What Does Article 5 of the AI Act Prohibit?

Article 5 of the AI Act prohibits placing on the market, putting into service or using AI systems for practices listed exhaustively in the ten lettered points of Article 5(1). The article has eight numbered paragraphs plus paragraphs 1a and 1b, and recitals 28 to 45 explain it, recital 28 giving the rationale of the prohibitions. The Commission calls this tier "unacceptable risk" (Guidelines C(2025) 5052, paragraph 2), a label of the Commission and not a term the act defines.

How does the act sort risk beyond the bans? The Commission groups AI systems into four tiers: unacceptable risk, high risk, transparency risk and minimal or no risk, and the overview of EU AI Act risk categories compares the four tiers with general-purpose AI models as a separate track.

Ten Lettered Points, Nine Practices: How the Bans Are Counted in the Current Text

Article 5(1), first subparagraph, has ten lettered points in its current text: (a), (b), (ba), (bb) and (c) to (h). The Commission policy page on the AI Act (last update 19 October 2026) states: "The AI Act prohibits nine practices". Both counts are right on their own terms, because the Commission joins points (ba) and (bb) into one practice, the ban on non-consensual sexually explicit and intimate content and child sexual abuse material.

Before 27 July 2026, Article 5(1) had eight lettered points, (a) to (h); that count is history. Regulation (EU) 2026/1744, Article 1, point (7), inserted points (ba) and (bb) and paragraphs 1a and 1b, and the consolidated text of 27 July 2026 shows the ten points in the act's order.

Every Ban in Article 5(1) at a Glance: Ten Points and the Pages That Explain Them

Ten bans make up Article 5(1), first subparagraph, each with its own conditions, operator verbs and carve-outs. The ten points read, in the act's order:

Article 5(1), first subparagraph, of Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744; carve-outs from the act and the non-binding guidelines C(2025) 5052.

Points (ba) and (bb) share one set of operator conditions in paragraph 1a and one exclusion in paragraph 1b, which is why the Commission counts them as one practice. Point (h) is the only ban with a procedure that permits an excepted use, set out in paragraphs 2 to 7.

Who the Bans Address: Providers, Deployers and Three Operator Verbs

The bans address providers and deployers through three operator verbs: placing on the market, putting into service and use. Points (a), (b), (ba), (bb) and (c) catch all three verbs, points (d) to (g) catch placing on the market and putting into service "for this specific purpose" plus use, and point (h) bans only the use, so selling a real-time remote biometric identification system is not prohibited (Guidelines C(2025) 5052, paragraph 11). The verb decides who breaches a ban: the provider who supplies a system or the deployer who uses it.

How wide is placing on the market? The guidelines read it to cover every means of supply, such as access through an API, cloud services, direct downloads, physical copies and embedding in physical products (paragraph 12). Putting into service covers supply for first use to third parties and in-house development and deployment (paragraph 13), so the provider vs deployer roles can sit in one organisation.

Open-source licences give no shelter: Article 2(12) excludes AI systems released under free and open-source licences unless they are placed on the market or put into service as high-risk AI systems or as AI systems that fall under Article 5 or 50, and the guidelines confirm that the exclusion does not reach prohibited practices (paragraph 36).

Section 2.5 of the guidelines discusses five scope exclusions of the act in relation to the bans: national security, defence and military purposes; cooperation with third-country authorities; research and development; purely personal non-professional activity; and free and open-source licences, the last of which does not reach Article 5. The page on AI Act exemptions explains the conditions of each Article 2 exclusion.

Where Prohibition Ends: Neighbouring Uses That Remain Lawful under Conditions

A use just outside an Article 5 point is usually not free of rules: the Commission guidelines state that exceptions to the bans mostly fall into the high-risk category (paragraph 37), and an Article 6(3) assessment never lifts a ban (paragraph 38). The boundary therefore runs between three outcomes: banned, high-risk under Annex III, or subject only to other duties of the act.

The boundary map pairs each point that has a lawful neighbour with the rule that takes over:

Article 5(1) of Regulation (EU) 2024/1689 as amended; boundaries from the non-binding guidelines C(2025) 5052, paragraphs 37 and 173 to 177.

Point (h) shows the boundary most clearly. Post-remote biometric identification is not an Article 5 practice at all; it is a high-risk use for which Article 26(10) requires the deployer, in a criminal investigation, to request authorisation ex ante or within 48 hours. Real-time use for one of the three objectives stays lawful only through the procedure of Article 5(2) to (5).

The guidelines also name uses outside point (c) and point (e). Scoring of legal entities, aggregated ratings, lawful credit scoring, fraud detection and proportionate telematics insurance pricing fall outside point (c) (paragraphs 173 to 177), while voice and other non-facial biometrics, fictitious faces and databases that are not expanded fall outside point (e) (paragraphs 234 to 236).

The 2026 Additions: Points (ba) and (bb) under Regulation (EU) 2026/1744

Regulation (EU) 2026/1744 added two bans to Article 5(1): point (ba) on realistic intimate or sexually explicit material of an identifiable person made without explicit consent, and point (bb) on child sexual abuse material; both apply from 2 December 2026. Point (ba) requires consent that is "freely-given, specific, informed, unambiguous and explicit". Point (bb) refers to Article 2, points (c) and (e), of Directive 2011/93/EU (OJ L 335, 17.12.2011, p. 1), except where a "without right" defence applies under national law.

Paragraph 1a sets two tests: a provider breaches the ban where generation is the intended purpose or a reasonably foreseeable and reproducible outcome without reasonable and adequate safeguards, and a deployer breaches it by using the system for that purpose. Paragraph 1b excludes manipulation that does not increase exposure of intimate parts or alter the nature of sexually explicit activities. The Commission proposal COM(2025) 836 did not contain point (ba); the co-legislators added it, and recital 11 of Regulation (EU) 2026/1744 refers to "nudification" applications.

When the Article 5 Bans Apply: Application Dates and the Start of Fines

Points (a), (b) and (c) to (h) and paragraphs 2 to 8 of Article 5 apply since 2 February 2025 (Article 113, third paragraph, point (a)); points (ba) and (bb) and paragraphs 1a and 1b apply from 2 December 2026. The penalties chapter followed on 2 August 2025 (Article 113, third paragraph, point (b)). Between those two dates the bans had direct effect without fines: national courts could enforce them, including through interim injunctions (Guidelines C(2025) 5052, paragraph 431).

Age gives no shelter from the bans. Only Article 111(1) and (2) apply "without prejudice to Article 5", so the prohibitions reach systems placed on the market before 2 February 2025 (Guidelines, paragraph 429).

On 2 December 2026 generative systems placed on the market before 2 August 2026 must also meet Article 50(2) under Article 111(4). The milestone page on the new Article 5 bans covers both obligations of that day.

What Providers and Deployers Must Do under Article 5

Article 5 obliges every provider and deployer to refrain from the listed practices, points (a), (b) and (c) to (h) since 2 February 2025 and points (ba) and (bb) from 2 December 2026; only point (h) and paragraph 1a attach positive conditions. The duties per operator read:

Article 5 of Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744.

No conformity route exists for a banned practice: in our reading, the provider removes the practice from the system or its intended purpose, and the deployer stops or never starts the use. Screening of this kind gives an orientation, not a legal assessment.

Fines and Enforcement for Prohibited AI Practices

Article 99(3) sets the maximum administrative fine that national penalty rules (Article 99(1)) provide for breaching an Article 5 ban: EUR 35 000 000 or, for an undertaking, 7 % of its total worldwide annual turnover for the preceding financial year, whichever is higher. The percentage applies only where the offender is an undertaking. EUR 35 million is the highest tier of the act, above the EUR 15 million tier for operator duties.

Chapter XII applies since 2 August 2025 (Article 113, third paragraph, point (b)).

National market surveillance authorities supervise the bans, and the AI Office is exclusively competent for AI systems based on a general-purpose AI model where model and system come from the same provider, except Annex I products, Annex III point 2 systems, systems of law enforcement, border management and financial institutions falling under Article 74(6), and Annex III point 8 administration of justice, and for AI systems that are or are integrated into very large online platforms or very large online search engines designated under the Digital Services Act (Article 75(1)). Article 75c(4) applies Article 99(3) to (7) mutatis mutandis to the AI Office, and the guide to fines of up to EUR 35 million sets out the full regime, including the national penalty rules. Each Member State decides to what extent fines can be imposed on public authorities and bodies (Article 99(8)).

Article 99(7) lists ten factors, points (a) to (j), for setting a fine, such as the nature, gravity and duration of the infringement, the size and market share of the operator, intent or negligence, and cooperation with the authorities. Recital 15 of Regulation (EU) 2026/1744 notes that criminal penalties are possible for the new Article 5 prohibitions.

Guidelines on Prohibited AI Practices: C(2025) 5052 and Its Limits

The Commission guidelines on prohibited AI practices explain each Article 5 point with examples; their formal reference is C(2025) 5052 final of 29 July 2025. Their content was approved by C(2025) 884 final of 4 February 2025 and published on the Commission library page the same day. Article 96(1)(b) is their legal basis (paragraph 3).

The guidelines run to 433 paragraphs in 12 sections, a 134-page PDF whose body text covers pages 1 to 129. They are non-binding: only the Court of Justice of the European Union gives an authoritative interpretation of the act (paragraph 5).

The guidelines predate Regulation (EU) 2026/1744 and do not cover points (ba) and (bb); the Commission library page shows no update for them as of 24 October 2026 (last update 31 July 2025). The register of EU AI Act guidelines lists these and every other Commission guideline under the act with its date and status.

Three studies on Article 5, published on the Commission library page on 8 May 2026 (last update 26 May 2026), add background that the same page describes as "only one input" for the Commission's guidelines:

• M.R. Leiser: points (a) and (b).
• E.J. Kindt: points (c), (d), (f) and (g) and paragraphs 3 to 8.
• C. Jasserand: points (h) and (e) and paragraph 2.

How the List of Prohibited Practices Is Reviewed: Article 112(1) and the 2026 Report

The Commission assesses once a year whether the list of prohibited practices in Article 5 needs amending (Article 112(1)). The duty runs until the end of the delegation period in Article 97 and covers Annex III as well. The first report, COM(2026) 234 final, is dated 20 May 2026 and was published on the Commission library page on 22 May 2026.

The report notes and further analyses a potential regulatory gap for AI systems generating child sexual abuse material and non-consensual intimate content, which points (a) and (b) do not cover (paragraph 57). It finds the assessment of other practices still at an early stage, defers a substantive evaluation until at least one year of application, and makes no formal amendment proposal (paragraph 61). Points (ba) and (bb) came through a separate route: the co-legislators added them to Regulation (EU) 2026/1744.

Member States, Union Law and the Reach of Article 5

Article 5 does not affect prohibitions that apply where an AI practice infringes other Union law (Article 5(8)), and recital 45 confirms that rule. Article 114 TFEU is the legal basis of the bans, with Article 16 TFEU added for remote biometric identification, biometric categorisation and individual risk assessment in law enforcement (Guidelines, paragraph 10; recital 3). That second legal basis explains why two Member States stand partly outside the article.

Ireland is not bound by Article 5(1), point (g), as it applies to police and judicial cooperation, points (d) and (h), Article 5(2) to (6) and Article 26(10), insofar as these rest on Article 16 TFEU (recital 40, Protocol No 21). Denmark is not bound by the same provisions under recital 41 and Protocol No 22, and the page on the AI Act in Ireland covers the Irish authorities under the act.

Article 5(5) leaves real-time remote biometric identification for law enforcement to national choice. A Member State permits it only by national law, notifies that law to the Commission within 30 days of adoption and can adopt stricter rules.

How to Screen an AI System for Article 5: Compliance Steps and Checklist

Five steps screen an AI system against Article 5: list the system and its intended purpose, test each of the ten points, check the carve-outs, record the result and re-check after changes. The steps run in this order:

1. List every AI system in development or use, with its intended purpose and the operator role.
2. Test each system against the ten points, (a), (b), (ba), (bb) and (c) to (h), and the operator verb that point uses.
3. Check the carve-outs and the Annex III counterpart of each point that comes close.
4. Record the result, the reasons and the date of the screening.
5. Repeat the screening after any change of intended purpose, model or deployment setting.

The guide to EU AI Act compliance sets out a step-by-step programme for providers and deployers that places Article 5 before the high-risk and transparency duties.

Generative systems need a second look before 2 December 2026. Their providers test the paragraph 1a safeguards for points (ba) and (bb), and their deployers check that no use serves that purpose.

The EU AI Act checklist sorts the obligations by role and risk class, with every deadline.

Prohibited AI Practices: Related Questions

Annex III, Article 2(1), the GDPR, Article 50, Articles 70 and 75 and Article 111 decide the questions that border the Article 5 bans.

Can a Use Close to a Ban Be a High-Risk AI System Instead?

Yes: uses next to points (d), (f), (g) and (h) can be high-risk under Annex III points 6(d), 1(c), 1(b) and 1(a).

The page on high-risk AI systems explains the Annex III counterparts and their duties. An Article 6(3) assessment never turns a banned practice into a lawful one.

Do Non-EU Providers Have to Respect the Bans?

Yes: Article 2(1), point (a), reaches providers placing AI systems on the Union market wherever they are established, and point (c) reaches third-country providers and deployers where the output is used in the Union.

The guide to the EU AI Act for non-EU companies explains how the act reaches non-EU providers. The bans apply to them exactly as to providers established in the Union.

How Does the GDPR Interact with Article 5?

The bans apply on top of data protection law, and Article 5(8) keeps every prohibition under other Union law intact.

The comparison AI Act vs GDPR shows where the two regulations overlap. Article 5(1), second subparagraph, makes point (h) without prejudice to Article 9 GDPR for biometric processing for purposes other than law enforcement.

Are Deepfakes Banned or Subject to Transparency Duties?

A deep fake, image, audio or video content under Article 3(60), is not banned as such; its deployer labels it under Article 50(4) since 2 August 2026.

The transparency obligations under Article 50 cover the labelling of deep fakes. Point (ba) bans non-consensual intimate material from 2 December 2026.

Which Authorities Enforce the Bans?

National market surveillance authorities enforce the bans; Article 70 required each Member State to designate a notifying authority, a market surveillance authority and a single point of contact by 2 August 2025.

The national competent authorities are listed for all 27 Member States. The AI Office acts instead for the Article 75(1) cases.

Are AI Systems Already on the Market Covered by the Bans?

Yes: the bans apply whatever the placing date of a system.

The page on AI systems already on the market explains the Article 111 transitional rules. Only Article 111(1) and (2) are "without prejudice to Article 5".

Frequently Asked Questions about Prohibited AI Practices

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

The ten lettered points of Article 5(1), points (a), (b) and (c) to (h) since 2 February 2025 and points (ba) and (bb) from 2 December 2026, ban harmful manipulation, exploiting vulnerabilities, non-consensual intimate material, child sexual abuse material, social scoring, crime prediction based solely on profiling, untargeted facial scraping, emotion inference at work and school, sensitive biometric categorisation and real-time remote identification for law enforcement.

How many AI practices does the AI Act prohibit?

Ten lettered points; the Commission counts nine practices because it joins (ba) and (bb). Before 27 July 2026 the act had eight points.

When did the prohibitions start to apply?

Points (a), (b) and (c) to (h) apply since 2 February 2025, with fines possible from 2 August 2025. Points (ba) and (bb) apply from 2 December 2026.

What is the maximum fine for a prohibited AI practice?

Article 99(3) sets the maximum that national penalty rules provide for: EUR 35 000 000 or, for an undertaking, 7 % of total worldwide annual turnover for the preceding financial year, whichever is higher. For SMEs and start-ups the lower amount applies.

Are the guidelines on prohibited AI practices binding?

No: C(2025) 5052 is non-binding; only the Court of Justice gives an authoritative interpretation.

What are examples of unacceptable-risk AI under the EU AI Act?

The Commission uses "unacceptable risk" for the Article 5 bans. Examples are social scoring that leads to detrimental treatment in unrelated contexts, untargeted scraping of facial images from the internet or CCTV footage, and inferring the emotions of employees at work.