Article 5 prohibits the placing on the market, the putting into service or the use of AI systems for the practices it lists, including manipulative techniques, exploitation of vulnerability, social scoring, certain predictive policing, untargeted facial image scraping, emotion inference at work and in education, biometric categorisation inferring sensitive attributes, and real-time remote biometric identification in publicly accessible spaces for law enforcement outside narrow exceptions. The original prohibitions apply from 2 February 2025. Two further prohibitions added by Regulation (EU) 2026/1744 apply from 2 December 2026. A practice that matches Article 5 is not a high-risk system to be managed; it is a practice that must not be placed on the market, put into service or used in the prohibited form.
What does Article 5 prohibit, in outline?
Article 5(1) lists prohibited practices in points (a) to (h), with points (ba) and (bb) added by the Omnibus amendments. The list covers techniques that manipulate behaviour or exploit vulnerability, social scoring that leads to detrimental treatment in unrelated contexts, predictive policing based solely on profiling or personality assessment, untargeted scraping of facial images to build or expand facial recognition databases, emotion inference in workplaces and education institutions, biometric categorisation that infers sensitive attributes, and real-time remote biometric identification in publicly accessible spaces for law enforcement. The two December 2026 additions cover AI systems generating or manipulating realistic intimate material of an identifiable person without qualifying consent, and material within Directive 2011/93/EU. Each point has its own wording, conditions and exceptions, and must be read separately.
How is a prohibition different from high-risk classification?
High-risk classification triggers a regime of requirements, documentation, conformity assessment and oversight. A prohibition does not trigger that regime for the prohibited practice; it forbids the practice in the form described. That difference changes the workflow. A team that finds a possible Article 5 match should stop and test the exact wording, including any exception, before investing in high-risk controls for the same use. A team that treats a prohibition as a risk to be mitigated by better documentation misunderstands the structure. The correct response to a confirmed match is to change the practice, remove the prohibited element, or not proceed, subject to qualified advice. The correct response to a near miss is a recorded screening decision that explains why the wording was not met.
Which dates apply to which prohibitions?
The original Article 5 prohibitions in points (a) to (h) apply from 2 February 2025, under Article 113 as originally set and as confirmed in the consolidated text. The new points (ba) and (bb) apply from 2 December 2026, following Regulation (EU) 2026/1744, which entered into force on 27 July 2026 and amended the application dates in Article 113. The December 2026 date is therefore not a general postponement or a grace period for the original prohibitions. A system that matched an original prohibition before that date was already within a prohibition that had begun to apply. Application dates should always be stated with their provision, and the consolidated text should be checked, because older summaries omit the December 2026 additions entirely.
What exceptions and conditions does Article 5 contain?
Article 5 is not a list of bare labels. Several points contain internal conditions, such as material distortion of behaviour and significant harm for manipulative techniques, exploitation of a specific vulnerability for the vulnerability prohibition, and solely profiling-based assessment for predictive policing. Emotion inference has a medical or safety exception. Biometric categorisation has a narrow carve-out for labelling or filtering lawfully acquired biometric data sets and for law enforcement categorisation in the circumstances stated. Real-time remote biometric identification has narrow law enforcement purposes with prior authorisation safeguards. The December 2026 points have their own provider and deployer conditions in new paragraphs (1a) and (1b), including safeguards and reasonably foreseeable misuse. Each exception must be tested on its own terms.
How should a screening record be kept?
A screening record should name the system, its intended purpose, the Article 5 point considered, the facts tested against each element of that point, the exception or condition considered, the conclusion, the reviewer, and the date. Where the conclusion is that a point is not met, the record should state which element failed, with evidence, so that a later reviewer can see the reasoning rather than a bare assertion. Where the conclusion is uncertain, the record should stay open, the system should not be treated as cleared, and qualified review should be sought. Screening should be repeated when the intended purpose, the user group, the data, or the deployment context changes, because a change in any of those can move a system into or out of a prohibition.
Continue with how the AI Act risk-based approach works.
What does this lead guide not decide?
This guide gives an outline map of Article 5. It does not decide whether any particular system, feature or use matches a prohibition, whether an exception applies, or what a specific organisation should do about a borderline practice. Those questions turn on the full wording of the relevant point, the facts of the use, and the safeguards in place, and they carry the highest stakes in the Act because the consequence is prohibition rather than a managed regime. Borderline cases should be confirmed with a competent authority or legal adviser, and the screening decision should be documented before the system is placed on the market, put into service or used.
Frequently asked questions
Are all Article 5 prohibitions already in force?
The original prohibitions in Article 5(1)(a) to (h) apply from 2 February 2025. The two prohibitions added as points (ba) and (bb) by Regulation (EU) 2026/1744 apply from 2 December 2026. The consolidated text and Article 113 should be checked for the provision in question.
Can a prohibited practice be made lawful by adding human oversight?
Not as a general rule. Article 5 prohibitions are not high-risk requirements that oversight can satisfy. Where an exception exists in the wording of a point, that exception must be met on its own terms; adding oversight does not create an exception that the text does not contain.
Does Article 5 apply to research prototypes?
Scope must be checked first under Article 2, including the research and development exclusions. A prototype that is placed on the market, put into service or used in the prohibited form does not escape Article 5 merely because it is called a prototype.
Where should screening sit in the workflow?
First, before high-risk classification. Article 5 is tested before Article 6, so that a prohibited practice is identified before effort is spent on the high-risk regime for the same use.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689 ↗AI Act Service Desk timeline ↗European Commission regulatory framework for AI ↗