LexLint.ioBeta

Know the law your code answers to.

Law / European Union

Commission Guidelines on prohibited AI practices under the AI Act (C(2025) 5052 final)

Commission Guidelines on prohibited artificial intelligence practices, C(2025) 5052 final

Guidance, not a law: the European Commission's reading of AI Act, Article 5 (prohibited AI practices). It binds nobody by itself; the law it reads does.

Guidance on an AI prohibited practices rule, addressed to public and private bodies.

Private right of action
No
Instrument type
guidance published by a regulator
How the hook was established
express
What makes it apply
Market targeting, Place of effect
Obligation class
Prohibition
Audit expectation
none

As of .

What the regulator expects

  • Do not place on the market, put into service or use an AI system that deploys subliminal, purposefully manipulative or deceptive techniques. The Guidelines read Article 5(1)(a) as met only where four conditions hold together: the placing on the market, putting into service or use of an AI system; the deployment of such a technique; a material distortion of behavior that appreciably impairs an informed decision; and significant harm caused or reasonably likely to be caused, with a plausible causal link between technique, distortion and harm (Section 3.2, paragraphs 60 and 61). Lawful persuasion, and manipulative or deceptive systems not reasonably likely to cause significant harm, fall outside it (Sections 3.5.1 and 3.5.2, paragraphs 127 and 134).NIST AI RMFMAP 3.3
  • Do not place on the market, put into service or use an AI system that exploits vulnerabilities due to age, disability or a specific social or economic situation. The Guidelines read Article 5(1)(b) as met only where four conditions hold together: the placing on the market, putting into service or use of an AI system; exploitation of those vulnerabilities; a material distortion of behavior; and significant harm caused or reasonably likely to be caused, with a plausible causal link (Section 3.3, paragraphs 98 and 99). Systems designed to protect and support people identified as vulnerable, such as banking services that use age or socio-economic situation in compliance with Union law, do not exploit vulnerabilities in that sense (Section 3.5.1, paragraph 133).NIST AI RMFMAP 3.3
  • Do not place on the market, put into service or use an AI system for the evaluation or classification of natural persons or groups over a certain period of time based on social behavior or personal or personality characteristics where the social score leads to detrimental or unfavorable treatment in unrelated social contexts or treatment that is unjustified or disproportionate. The Guidelines read Article 5(1)(c) as met only where three conditions hold together (Section 4.2, paragraphs 149 and 150). Scoring of legal entities falls outside it in principle, and lawful evaluation practices carried out for a specific purpose in accordance with Union and national law are not prohibited per se (Section 4.3, paragraphs 173 and 175).NIST AI RMFMAP 3.3
  • Do not use an AI system for risk assessments of natural persons to assess or predict the risk of committing a criminal offence based solely on profiling or on an assessment of personality traits and characteristics. The Guidelines read Article 5(1)(d) as met only where three conditions hold together (Section 5.2, paragraphs 187 and 188). Location-based or place-based crime prediction, which in principle assesses no specific individual, and AI systems that support a human assessment already based on objective and verifiable facts directly linked to a criminal activity fall outside it (Sections 5.3.1 and 5.3.2, paragraphs 212 and 214).NIST AI RMFMAP 3.3
  • Do not place on the market or put into service for this specific purpose, or use, an AI system that creates or expands facial recognition databases through untargeted scraping of facial images from the internet or CCTV footage. The Guidelines read Article 5(1)(e) as met only where four conditions hold together, and as reaching tools for untargeted scraping, not every scraping tool; scraping is untargeted where it operates without a specific focus on a given individual or group, and respecting robots.txt does not change that (Section 6.2, paragraphs 224, 225 and 228). Scraping of biometric data other than facial images, scraping without an AI system, facial image databases not used for the recognition of persons, and models that generate images of fictitious persons fall outside it (Section 6.3, paragraphs 234 and 235).NIST AI RMFMAP 3.3
  • Do not place on the market or put into service for this specific purpose, or use, an AI system to infer emotions of a natural person in the areas of workplace and education institutions, except where the use is intended for medical or safety reasons. The Guidelines read Article 5(1)(f) as met only where four conditions hold together (Section 7.2, paragraphs 242 and 243), and read the medical and safety exception narrowly: detecting general aspects of wellbeing, such as stress or burnout, is not covered, and safety means the protection of life and health (Section 7.2.3, paragraphs 256 to 258). Systems that infer emotions without using biometric data, and systems that infer physical states such as pain or fatigue, fall outside it (Section 7.4, paragraph 265).NIST AI RMFMAP 3.3
  • Do not place on the market or put into service for this specific purpose, or use, a biometric categorization system that categorizes individual 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. The Guidelines read Article 5(1)(g) as met only where five conditions hold together (Section 8.2, paragraphs 273 and 274). The labeling or filtering of lawfully acquired biometric datasets, including in the area of law enforcement, falls outside it (Section 8.3, paragraphs 284 and 286).NIST AI RMFMAP 3.3
  • Do not use a real-time remote biometric identification system in publicly accessible spaces for law enforcement purposes. The Guidelines read Article 5(1)(h) as met where the system is a remote biometric identification system, the activity is its use, in real time, in publicly accessible spaces, for law enforcement purposes (Section 9.2, paragraph 295). The three exceptions are exhaustive and are not themselves a legal basis, so in the absence of Member State legislation authorizing real-time use the use is prohibited (Section 9.3, paragraph 326). Retrospective use, biometric verification, real-time use in a private or online space, and use by private actors fall outside it (Section 10.7, paragraph 426).NIST AI RMFMAP 3.3
  • Where national law permits real-time remote biometric identification for law enforcement under an Article 5(1)(h) exception, do not use the system until a fundamental rights impact assessment has been completed and the competent judicial or independent administrative authority has authorized the use, except in a duly justified urgent case; register the system in the EU database, without undue delay in a justified emergency; and notify each use to the market surveillance authority and the national data protection authority (Section 10, paragraphs 377, 380, 382, 383 and 409).NIST AI RMFMAP 5.1NIST AI RMFGOVERN 1.1
  • Check whether an exclusion in Article 2 applies before treating Article 5 as engaged, reading each as the Guidelines do: military, defense or national security only where the system is used exclusively for those purposes (Section 2.5.1, paragraph 24); research, testing or development only before the system is placed on the market or put into service, with real-world testing not covered (Section 2.5.3, paragraphs 30 and 32); personal non-professional activity only for the obligations of natural-person deployers (Section 2.5.4, paragraphs 34 and 35); and no exclusion for a free and open-source system that falls under Article 5 (Section 2.5.5, paragraph 36).
  • Providers of AI systems, including general-purpose AI systems, are expected not to place on the market or put into service a system reasonably likely to behave or be directly used in a manner prohibited by Article 5, and to take effective and verifiable safeguards to the extent reasonably foreseeable, feasible and proportionate; deployers are expected not to use any AI system in a prohibited manner, including by bypassing the provider's safety guardrails (Section 2.7, paragraphs 39 to 41).NIST AI RMFMAP 3.3NIST AI RMFMANAGE 1.3
  • Comply with the other Union law that continues to apply alongside Article 5, since respecting the prohibitions is not a sufficient condition for compliance with it: the General Data Protection Regulation, the Law Enforcement Directive and Regulation (EU) 2018/1725 (paragraph 45), with the AI Act applying as lex specialis to Article 10 of the Law Enforcement Directive for biometric categorization and real-time remote biometric identification used for law enforcement (paragraph 46), and Union consumer protection and safety legislation (paragraph 47) (Section 2.8, paragraphs 43 to 47).NIST AI RMFGOVERN 1.1

In the frameworks

NIST AI RMF
  • GOVERN 1.1Legal and regulatory requirements involving AI are understood, managed, and documented.
  • MAP 3.3Targeted application scope is specified and documented based on the system’s capability, established context, and AI system categorization.
  • MAP 5.1Likelihood and magnitude of each identified impact (both potentially beneficial and harmful) based on expected use, past uses of AI systems in similar contexts, public incident reports, feedback from those external to the team that developed or deployed the AI system, or other data are identified and documented.
  • MANAGE 1.3Responses to the AI risks deemed high priority, as identified by the MAP function, are developed, planned, and documented. Risk response options can include mitigating, transferring, avoiding, or accepting.
NIST AI 600-1
MIT mitigations Draft
  • 1.5Safety Decision Frameworks
  • 3.3Access Management
NIST Privacy Framework
  • GV.PO-P5Legal, regulatory, and contractual requirements regarding privacy are understood and managed.
  • CT.PO-P1Policies, processes, and procedures for authorizing data processing (e.g., organizational decisions, individual consent), revoking authorizations, and maintaining authorizations are established and in place.
FINOS AIGF
ETSI EN 304 223
  • 5.1.2-1Business need and security risks written down and assessed before deciding to build an AI system.
  • 5.1.4-5End-users told which uses of the system are prohibited.

Our reading of where this law's duties sit in each framework, and in the NIST AI RMF, line by line beside each duty above. It never says that running a control meets the law. How we read a framework against the law

Identifiers refer to ETSI EN 304 223 V2.1.1 (2025-12), © ETSI 2025, all rights reserved. ETSI owns the standard; no ETSI text (titles or sentences) is reproduced here. The standard is free to download from ETSI.

Who enforces it

Enforcement body

The market surveillance authorities designated by the Member States, and the European Data Protection Supervisor for the EU institutions, agencies and bodies, which enforce the AI Act's rules for AI systems, including the prohibitions (paragraph 53).

Settledness

Paragraph 5 says the Guidelines are non-binding and leaves any authoritative interpretation of the AI Act to the Court of Justice of the European Union.

As of
Open questions
Which of the Guidelines' readings of the Article 5 prohibitions will the Commission amend or withdraw as market surveillance enforcement and Court of Justice rulings supply experience, given that paragraph 433 reserves the power to do so?

What it makes you log

Who may demand the log

Regulator

Logging duty

Paragraph 409 reads Article 5(4) as requiring each use of a real-time remote biometric identification system in publicly accessible spaces for law enforcement to be notified to the market surveillance authority and the national data protection authority.

Kind
Implicit
As of
Provision
Section 10.3, paragraph 409

What this law does

Drafted with AI

Drafted with AI from the cited sources under the direction of UnGovr staff. UnGovr holds editorial responsibility for this page. How this site is made

Research summary

Legal information, not legal advice. This is LexLint's own research summary of a public legal source, and it creates no attorney-client relationship. For decisions that matter, consult qualified counsel in the relevant jurisdiction. About LexLint

Paragraph 5 says that the Guidelines are non-binding and that any authoritative interpretation of the AI Act may ultimately only be given by the Court of Justice of the European Union. Paragraph 3 says that, pursuant to Article 96(1)(b) of the AI Act, the Commission is to adopt guidelines on the practical implementation of the practices prohibited under Article 5.

Paragraph 4 says that the Guidelines aim to increase legal clarity and to provide insights into the Commission’s interpretation of the prohibitions in Article 5 of the AI Act. The Communication to the Commission of , reference C(2025) 884 final, approves the content of the draft Guidelines. The approval Communication of says that the Guidelines will be formally adopted by the Commission at a later date, when all language versions are available.

The approval Communication of says that it is only from the moment of formal adoption that the Guidelines will be applicable. The Guidelines text is a Communication from the Commission dated , reference C(2025) 5052 final. Paragraph 433 says that the Commission will review the Guidelines as soon as this is necessary in view of practical experience gained in the implementation of the prohibitions and the pace of technological, societal, and regulatory developments.

Paragraph 433 also says that, during such a review, the Commission may decide to withdraw or amend the Guidelines. Paragraph 7 says that the application of Article 5 of the AI Act will require a case-by-case assessment and that the examples in the Guidelines are merely indicative. Paragraph 429 says that Article 5 of the AI Act applies from and in principle to all AI systems, whether they were placed on the market or put into service before or after that date.

Paragraph 430 says that the provisions on penalties for non-compliance with the prohibitions in Article 5 of the AI Act will not apply before . Paragraph 431 says that the prohibitions have direct effect and thus enable affected parties to enforce them in national courts and request interim injunctions against the prohibited practices.

Paragraph 53 says that market surveillance authorities designated by the Member States and the European Data Protection Supervisor are responsible for the enforcement of the rules in the AI Act for AI systems, including the prohibitions. Paragraph 55 says that providers and deployers engaging in prohibited AI practices may be fined 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.

Paragraph 57 says that, since violations of the prohibitions in Article 5 of the AI Act interfere the most with the freedoms of others and give rise to the highest fines, their scope should be interpreted narrowly. Paragraph 24 says that the exclusion for military, defense or national security purposes in Article 2(3) of the AI Act applies only where the AI system is placed on the market, put into service or used exclusively for those purposes.

Paragraph 30 says that Article 2(8) of the AI Act excludes any research, testing or development activity regarding AI systems or AI models prior to their being placed on the market or put into service. Paragraph 32 says that the Article 2(8) exclusion is without prejudice to the obligation to comply with the AI Act where an AI system is placed on the market or put into service as a result of research and development activity.

Paragraph 34 says that Article 2(10) of the AI Act excludes the obligations of deployers who are natural persons using systems in the course of a purely personal non-professional activity. Paragraph 36 says that providers of AI systems cannot benefit from the free and open-source licence exclusion if the AI system they place on the market or put into service constitutes a prohibited practice under Article 5 of the AI Act.

Paragraph 14 says that the reference to use in Article 5 of the AI Act includes any misuse of an AI system, reasonably foreseeable or not, that may amount to a prohibited practice. Paragraph 11 says that the prohibition in Article 5(1)(h) of the AI Act applies only to the use of real-time remote biometric identification systems.

Paragraph 40 says that providers also have a responsibility not to place on the market or put into service AI systems, including general-purpose AI systems, that are reasonably likely to behave or be directly used in a manner prohibited by Article 5 of the AI Act. Paragraph 40 gives the example of a general-purpose AI system used as a chatbot that may deploy manipulative and deceptive techniques which is likely to cause significant harm.

Paragraph 43 says that respecting the prohibitions in the AI Act is not a sufficient condition for compliance with other Union legislation that remains applicable to providers and deployers of AI systems. Paragraph 45 says that the General Data Protection Regulation, the Law Enforcement Directive and Regulation (EU) 2018/1725 remain unaffected under Article 2(7) of the AI Act and continue to apply alongside it.

Paragraph 46 says that, for biometric categorisation systems and real-time remote biometric identification systems used for law enforcement, the AI Act applies as lex specialis to Article 10 of the Law Enforcement Directive.

Paragraph 149 says that, under Article 5(1)(c) of the AI Act, a social score must lead or be capable of leading to detrimental or unfavorable treatment of persons or groups, in social contexts unrelated to those in which the data was originally generated or collected or in a way that is unjustified or disproportionate to their social behavior or its gravity.

Paragraph 227 says that scraping typically refers to using web crawlers, bots, or other means to extract data or content from different sources, including CCTV, websites or social media, automatically. Paragraph 225 says that the prohibition in Article 5(1)(e) of the AI Act applies to scraping tools placed on the market or put into service for the specific purpose of untargeted scraping of facial images from the internet or CCTV footage.

Paragraph 273 says that the prohibition in Article 5(1)(g) of the AI Act requires a biometric categorization system that categorizes individual 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.

Paragraph 326 says that, in the absence of Member State legislation authorizing the use of real-time remote biometric identification for one or more of the objectives in Article 5(1)(h), such use is prohibited from . The overview table in paragraph 9 of the Guidelines lists the prohibitions of Article 5(1) as points (a) to (h).

Back to the example  ·  Lint your app