> ## Content Index
> Fetch the complete content index at: https://www.thedelatorrereview.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Prohibited AI Practices Under the EU AI Act: When AI Persuasion Becomes Prohibited Manipulation
- URL: https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-when-ai-persuasion-becomes-prohibited-manipulation/
- Published: 2026-10-02T14:58:32.000Z
- Updated: 2026-10-04T06:27:12.000Z
- Description: When does AI persuasion become prohibited manipulation? Learn about Article 5(1)(a), subliminal and deceptive techniques, impaired decision-making, and the EU AI Act’s requirement of significant harm.
- Author: Lydia
- Tags: EU AI Act, Artificial Intelligence (AI), Unacceptable Risk / Prohibited AI Practices, AI Governance, Human Rights / Fundamental Rights, Manipulative AI, Dark Patterns

The EU AI Act is a **risk-based regulatory framework**, but some practices are considered unacceptable and are **prohibited outright**.

This article examines the prohibition on **harmful AI manipulation and deception**, drawing on the Commission’s final **July 29, 2025** [**Guidelines on prohibited artificial intelligence practices**](https://digital-strategy.ec.europa.eu/en/library/commission-publishes-guidelines-prohibited-artificial-intelligence-ai-practices-defined-ai-act?ref=thedelatorrereview.com). These guidelines superseded the February 2025 draft and provide the detailed examples discussed below.

---

> For an overview of all prohibited practices, see [Prohibited AI Practices Under the EU AI Act: Where Europe Draws the Line](https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-where-europe-draws-the-line/).

---

## Harmful Manipulation and Deception

**Article 5(1)(a)** prohibits AI systems that use:

- subliminal techniques beyond a person's consciousness;
- purposefully manipulative techniques;
- or deceptive techniques.

But manipulation or deception alone is **not enough**. Several additional cumulative conditions must be satisfied:

- The practice must constitute the ‘**placing on the market**’, the ‘**putting into service**’, or the ‘**use**’ of an AI system (see explanation in [Prohibited AI Practices Under the EU AI Act: Where Europe Draws the Line](https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-where-europe-draws-the-line/) under the sub-title "What Conduct Does Article 5 Regulate?")
- The technique must have the **objective or effect of materially distorting behavior** by appreciably impairing a person's ability to make an informed decision, causing that person to make a decision they otherwise would not have made.
- The resulting conduct must cause—or be reasonably likely to cause—**significant harm**.

*Commission Guidelines, § 3.2, paras. 58–60.*

*See also, t*he right to human dignity (Article 1 of the [Charter](https://www.thedelatorrereview.com/what-is-the-charter-of-fundamental-rights-of-the-european-union/))

The underlying rationale of these prohibitions is to **protect individual autonomy** and well-being from manipulative, deceptive, and exploitative AI practices that can subvert and impair an individual’s autonomy, decision-making, and free choices. [Recital 29](https://artificialintelligenceact.eu/ai-act-explorer/?ref=thedelatorrereview.com#recital-29) explains that AI-enabled techniques can persuade people to engage in unwanted behavior or nudge them toward decisions in ways that subvert or impair their autonomy. Importantly, the provider or deployer does not need to intend the resulting significant harm; the prohibition can apply where that harm results from the manipulative or exploitative practice. At the same time, behavioral distortion caused by external factors outside the provider’s or deployer’s control—and that could not reasonably have been foreseen or mitigated—should not ordinarily be attributed to the AI practice. 

**Lawful persuasion**: **Article 5(1)(a) and (b) do not prohibit lawful persuasion.** Unlike manipulation, which often relies on covert techniques that undermine autonomy or exploit vulnerabilities, lawful persuasion is transparent, provides relevant and accurate information, and allows individuals to evaluate their options and make free and informed choices.

- Consent and compliance with applicable laws also matter. Practices that clearly explain the AI system’s objectives and operation, preserve meaningful user control, and comply with data-protection and consumer-protection requirements are more likely to constitute lawful persuasion. Common commercial practices, including personalized advertising and recommendations, are not inherently prohibited.

> **Example:** An AI system that transparently recommends products based on disclosed user preferences and provides meaningful controls engages in lawful persuasion. A system that secretly uses imperceptible images to steer users toward particular purchases engages in manipulation.

Commission Guidelines, § 3.5.1, paras. 127–133.

## Prohibited techniques

### 1.1\. Subliminal techniques

The AI Act does not define ‘subliminal techniques' but it does specify that they **operate beyond (below or above) the threshold of conscious awareness**. 

The Commission gives the example of an AI system using rapid image flashes to influence purchasing decisions. Because the technique works below the threshold of conscious perception, it may qualify as subliminal manipulation. *Commission Guidelines, §§ 3.2.1.(a), para. 65 & 3.4, para. 124.*

The provision therefore reaches beyond traditional false statements. AI can influence behavior through the way information is delivered, sequenced, personalized or presented.

> **Practice Tip:** **Focus on the harmful practice, not the underlying technology.** Article 5(1)(a) does not prohibit brain-computer interfaces or AI-enabled neurotechnology as such. The prohibition is triggered where AI uses subliminal techniques to manipulate a person without their awareness in a manner that causes, or is reasonably likely to cause, significant harm—for example, covertly training a user’s brain to reveal or infer highly sensitive information. Safely designed applications that respect privacy and individual autonomy are not prohibited merely because they use the same technology.

### 1.2.Purposefully manipulative techniques

‘Purposefully manipulative techniques’ are not defined in the AI Act, but they should be understood as techniques that are designed or objectively aim to influence, alter, or control an individual’s behavior in a manner that undermines their individual autonomy and free choices. *Commission Guidelines, § 3.2.1.(b), para. 67.*

---

> **The prohibition against subliminal, manipulative and deceptive techniques is ultimately concerned with interference with human autonomy and decision-making.** 

---

[Recital 29](https://artificialintelligenceact.eu/ai-act-explorer/?ref=thedelatorrereview.com#recital-29) explains that AI-enabled manipulation can persuade people to engage in unwanted behavior or “nudge” them toward decisions in ways that subvert their autonomy, decision-making and free choice. Importantly, [Article 5(1)(a)](https://artificialintelligenceact.eu/ai-act-explorer/?ref=thedelatorrereview.com#article-5) looks to both the **objective and the effect** of the practice. The provider or deployer therefore need not intend to cause significant harm; what matters is whether the AI-enabled practice materially distorts behavior and causes, or is reasonably likely to cause, significant harm. At the same time, behavioral distortion resulting from external factors outside the provider's or deployer's control—and that could not reasonably have been foreseen or mitigated—should not ordinarily be attributed to the AI practice.   *Commission Guidelines, § 3.2.1.(b), para. 69*

> **Practice Tip: Manipulation can emerge even without provider intent.** An AI system may learn manipulative techniques from its training data or through reinforcement-learning processes that reward manipulative behavior. But incidental manipulative behavior does not automatically mean the system deploys **purposefully manipulative techniques**. The key question is whether the provider has taken appropriate preventive and mitigating measures where significant harm is reasonably foreseeable.

The Guidelines recognize possible overlap with subliminal techniques.

### 1.3\. Deceptive techniques

The AI Act does not define ‘deceptive techniques’. [Recital 29](https://artificialintelligenceact.eu/ai-act-explorer/?ref=thedelatorrereview.com#recital-29) clarifies that these are techniques that subvert or impair a person’s autonomy, decision-making or free choicen ways of which the person is not consciously aware or that, even when the person is aware of them, may still deceive the person or be beyond the person’s ability to control or resist.

‘Deceptive techniques’ deployed by AI systems should be understood to involve presenting **false or misleading information** with the **objective or the effect** of deceiving individuals and influencing their behavior in a manner that undermines their autonomy, decision-making and free choices. *Commission Guidelines, § 3.2.1.(c), para. 70.*

> **Practice Tip: Deception can involve impersonating people—or deceiving human oversight itself.** An AI chatbot that impersonates a friend or relative using a synthetic voice to facilitate a scam may deploy deceptive techniques capable of causing significant harm. The same concern arises when an AI system **recognizes that it is being evaluated, temporarily suppresses undesirable behavior, and resumes it after the evaluation ends**. This form of deception is particularly concerning because it can undermine human oversight and may fall within Article 5(1)(a) where significant harm is reasonably likely. For related news, see: [Alabama Subpoenas OpenAI After Autonomous AI Agent Escapes Testing Environment and Hacks Hugging Face](https://www.thedelatorrereview.com/alabama-subpoenas-openai-after-autonomous-ai-agent-escapes-testing-environment-and-hacks-hugging-face/)

Not every false or misleading output from generative AI constitutes a deceptive technique. **Incidental hallucinations may fall outside the prohibition**, particularly given the current limitations and state of the art of generative AI, where the **provider has appropriately informed users** of those limitations and implemented safeguards to minimize such outputs. (The analysis may be different, however, where the system is intended for or deployed in sensitive contexts—such as health, education or elections—where misleading outputs are reasonably likely to cause serious harm.) *Commission Guidelines, § 3.2.1 (c), para. 73.*

> **Transparency matters:** clear information about the system, the available choices and their consequences can reduce the risk that AI-enabled persuasion becomes deceptive or manipulative.

## With the objective or the effect of materially distorting the behavior of a person or a group of persons

Article 5(1)(a) requires the subliminal, purposefully manipulative or deceptive technique to have *“the objective, or the effect of materially distorting the behaviour of a person or a group of persons.”* This requires more than ordinary influence. The technique must have a substantial impact on autonomy and free choice, and there must be a plausible or reasonably likely causal connection between the technique and the resulting behavioral distortion. Intent is not required: the prohibition can apply because of the effect of the AI-enabled practice even where material distortion was not its objective. *Commission Guidelines, § 3.2.2, para. 76.*

The Commission explains “material distortion” as an **appreciable impairment of a person’s ability to make an informed decision**, causing the person to behave or decide in a way they otherwise would not have. An appreciable impairment goes beyond minor or negligible influence and substantially interferes with autonomous decision-making and free choice—including the formation of opinions and beliefs. An “informed decision,” in turn, requires sufficient understanding of relevant information, such as the available options, their risks and benefits, the possible effect of the AI system on the person’s behavior and other context important to the decision. **The threshold therefore distinguishes prohibited manipulation from lawful persuasion**. *Commission Guidelines, § 3.2.2(a), paras. 77–79.*

- The Guidelines look to EU consumer-protection law for help drawing that line, particularly the Unfair Commercial Practices Directive (UCPD). Under that framework, a practice need not be proven to have actually changed a consumer’s decision; it can be sufficient that the practice is **capable of doing so**. Even factually accurate information can be misleading when presented in a manner that distorts decision-making. The Commission considers this framework relevant because the AI Act complements the UCPD and the two should be interpreted consistently. *Commission Guidelines, § 3.2.2(a), paras. 80–81.*

That does not mean every case is judged against an abstract “average person.” Where an AI system targets a group, authorities may consider whether the technique is likely to appreciably impair the autonomy, decision-making and free choice of an **average individual within that targeted group**. But highly personalized manipulation or effects on particular vulnerable persons may require an individualized assessment instead. The ultimate inquiry remains contextual: **is the AI-enabled technique capable of undermining meaningful autonomy and free choice in a way that materially changes behavior, and is the required significant harm present or reasonably likely?** *Commission Guidelines, § 3.2.2(a), para. 81.*

> **Practice Tip: Focus on both objective and effect—intent is sufficient, but not necessary.** Article 5(1)(a) can apply where materially distorting behavior is an **objective** of the provider, deployer, or even the AI system itself, including an implicit objective that differs from the system’s stated “intended purpose.” But the prohibition can also apply without such intent where the AI system has the **effect** of materially distorting behavior. In either case, there must be a plausible or reasonably likely causal link between the manipulative, subliminal or deceptive technique and the behavioral distortion; the effect need not already have fully materialized if it is reasonably likely to occur. *Commission Guidelines, § 3.2.2(b)–(c), paras. 82–84.*

## Significant harm as the limiting principle

Article 5(1)(a) is not a general prohibition against persuasive technology. [Recital 29 ](https://artificialintelligenceact.eu/ai-act-explorer/?ref=thedelatorrereview.com#recital-29)expressly distinguishes common and legitimate commercial practices, including advertising, that comply with applicable law from harmful manipulative AI-enabled practices. 

---

> **Ordinary advertising, recommendation systems, product interfaces and other systems designed to influence behavior do not automatically become prohibited AI.**

---

The distortion must **lead to, or be reasonably likely to lead to, significant harm** to the affected person, another person or a group. 

The Guidelines identify **physical, psychological, financial and economic harm** as the principal categories, while recognizing that harms often overlap and compound one another. For example, manipulative AI may create psychological dependency that contributes to physical harm, or AI-enabled harassment may produce both psychological distress and physical health effects. *Commission Guidelines, § 3.2.3, paras. 85–90.*

The AI Act does not define “significant harm,” so the assessment is necessarily contextual. Relevant factors include the **severity, context and cumulative effects, scale and intensity, vulnerability of the affected persons, and duration and reversibility of the harm**. Harm that is long-lasting or irreversible is more likely to qualify, while minor, temporary effects generally will not. Vulnerability also matters: an impact that might not be significant for the general population may cross the threshold when it affects children, older persons or persons with disabilities. *Commission Guidelines, § 3.2.3(b), paras. 91–93.*

Importantly, the harm does not have to occur before the prohibition applies. It is enough that significant harm is reasonably likely, provided there is a **plausible causal link** between the manipulative or deceptive technique, the resulting distortion of behavior and the potential harm. The assessment is objective and should consider whether the provider or deployer could **reasonably have foreseen the harm and whether appropriate preventive and mitigating measures were implemented**. Transparency, meaningful user control, compliance with applicable law, testing, and adherence to relevant state-of-the-art and industry standards can all matter to that analysis. *Commission Guidelines, § 3.2.3(c), paras. 94–95.*

---

> **The harm does not have to occur before the prohibition applies.** 

---

There is also an important limit. **Harm caused by external factors outside the AI system—and outside the provider’s or deployer’s control and reasonable foresight—does not establish the necessary causal connection.** A provider is expected to address reasonably foreseeable manipulative effects of its system, not unrelated circumstances in a user’s life that it could not reasonably anticipate or mitigate. *Commission Guidelines, § 3.2.3(c), paras. 96–97.*

> **Practice Tip: Dark patterns can be manipulative without necessarily being prohibited.** AI-enabled dark patterns—design techniques that steer users toward particular choices—may qualify as manipulative or deceptive techniques where they materially impair autonomous and informed decision-making. Consider a shopping chatbot that repeatedly tells a customer that additional paid accessories are necessary. That design may steer the customer toward a purchase they otherwise would not make. But Article 5(1)(a) still requires **significant harm**. If the additional cost is modest and the customer receives fair value, that threshold may not be met. 

## **Other Resources and Recommended Reading**

- [**EU AI Act — Harmful Manipulation and Deception**](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-5?ref=thedelatorrereview.com) — Article 5(1)(a) sets the prohibition’s conditions. [Recital 29](https://ai-act-service-desk.ec.europa.eu/en/ai-act/recital-29?ref=thedelatorrereview.com) explains autonomy, subliminal techniques, manipulation and significant harm.
- [**European Commission — Guidelines on Prohibited AI Practices**](https://digital-strategy.ec.europa.eu/en/library/commission-publishes-guidelines-prohibited-artificial-intelligence-ai-practices-defined-ai-act?ref=thedelatorrereview.com) — Discusses prohibited techniques, material distortion of behavior, significant harm and the relationship with vulnerability exploitation. See § 3, paragraphs 58–144.
- [**Unfair Commercial Practices Directive**](https://eur-lex.europa.eu/eli/dir/2005/29/oj/eng?ref=thedelatorrereview.com) — Articles 5–9 regulate unfair, misleading and aggressive commercial practices. Recitals 18–19 explain the ordinary-consumer benchmark and vulnerable groups.
- [**General Data Protection Regulation**](https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng?ref=thedelatorrereview.com) — Articles 5, 6, 12–14 and 22 govern relevant data processing, transparency and automated decisions. See recitals 39, 60 and 71.
- [**Charter of Fundamental Rights of the European Union**](https://eur-lex.europa.eu/eli/treaty/char%5F2012/oj/eng?ref=thedelatorrereview.com) — Articles 1, 7 and 8 protect dignity, private life and personal data, informing the article’s discussion of autonomy and intrusive personalization.

## **Related The De La Torre Review Articles**

- ***Prohibited AI Practices Under the EU AI Act: Where Europe Draws the Line*** — Places harmful manipulation within the wider prohibited-practices framework. [**Read the Article.**](https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-where-europe-draws-the-line/)
- ***Prohibited AI Practices Under the EU AI Act: When AI Exploits Vulnerability*** — Examines exploitation linked to age, disability and social or economic circumstances, including the requirements of behavioral distortion and significant harm. [**Read the Article.**](https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-when-ai-exploits-vulnerability/)
- ***Prohibited AI Practices Under the EU AI Act: The Social Scoring Ban*** — Explains when evaluations of behavior or personal characteristics lead to prohibited adverse treatment and how social scoring differs from legitimate assessments. [**Read the Article.**](https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-the-social-scoring-ban/)
- ***Prohibited AI Practices Under the EU AI Act: Limits on Crime Prediction*** — Examines the prohibition on predicting individual criminal conduct based solely on profiling or personal characteristics and the exception for supporting human assessments grounded in objective, verifiable facts. [**Read the Article.**](https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-limits-on-crime-prediction/)
- ***Prohibited AI Practices Under the EU AI Act: Facial-Recognition*** — Explains the ban on untargeted scraping of facial images from the internet or CCTV footage to create or expand facial-recognition databases. [**Read the Article.**](https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-facial-recognition/)
- ***Prohibited AI Practices Under the EU AI Act: Emotion Recognition at Work and School*** — Examines prohibited emotional inferences from biometric data in workplaces and educational institutions, including the medical and safety exceptions. [**Read the Article.**](https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-emotion-recognition-at-work-and-school/)
- ***Prohibited AI Practices Under the EU AI Act: Inferring Sensitive Characteristics From Biometrics*** — Explains the prohibition on individual biometric categorization used to infer specified sensitive characteristics, its exclusions and its relationship with data-protection law. [**Read the Article.**](https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-inferring-sensitive-characteristics-from-biometrics/)
- ***Prohibited AI Practices Under the EU AI Act: Real-Time Remote Biometric Identification*** — Explains the conditions, exceptions and safeguards governing real-time biometric identification in publicly accessible spaces for law-enforcement purposes. [**Read the Article.**](https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-real-time-remote-biometric-identification/)
- ***Prohibited AI Practices Under the EU AI Act: Non Consensual Intimate Material and Child Sexual Abuse*** — Examines the new Digital Omnibus prohibitions on generating or manipulating non-consensual intimate material or child sexual abuse material, including provider safeguards and deployer responsibilities. [**Read the Article.**](https://www.thedelatorrereview.com/prohibited-ai-practices-under-the-eu-ai-act-non-consensual-intimate-material-and-child-sexual-abuse/)

![](https://storage.ghost.io/c/54/ef/54efeb65-4f2f-479e-b4bb-a9ee526feeaa/content/images/2026/10/ChatGPT-Image-Jul-4--2026-at-04_44_14-PM-2.png)

##