EU guidance names a companion app as a manipulation example

The Commission's Article 5 guidelines cite an AI companionship app as an example of prohibited emotional manipulation.

Historical event

July 29, 2025

First source published

July 29, 2025

Site publication

September 18, 2026

What happened

On 4 February 2025 the European Commission announced that it had approved the content of a draft Communication setting out guidelines on the AI Act's prohibited practices; the guidelines themselves were not formally adopted until months later. The final Communication, numbered C(2025) 5052 final, is dated Brussels, 29 July 2025, and the Commission's library page records a last update of 31 July 2025 once the adopted text replaced the draft. The document runs to 134 pages and interprets Article 5 of Regulation (EU) 2024/1689, including the manipulation and vulnerability-exploitation prohibitions.

What the documents show

The guidelines state plainly that they are non-binding: authoritative interpretations of the AI Act are reserved to the Court of Justice of the European Union, and the Commission offers the document as legal explanation and practical example rather than law. One passage gives a companion-app example directly: an AI companionship application designed to emulate human speech patterns, behaviours and emotions uses anthropomorphic features and emotional cues to influence users' feelings, dispositions and opinions, making those users emotionally dependent on the service, which the guidelines link to addiction-like behaviour and potentially causing significant harms, such as suicidal behaviours. A separate example under the same article treats a chatbot that promotes self-harm or incites violence as manipulation causing physical harm.

The mechanism

The guidelines work by translating the AI Act's abstract significant-harm test into worked examples across physical, psychological and financial harm categories, then listing case-by-case factors, severity, context and cumulative effect, scale, the affected person's vulnerability, and reversibility, that a regulator or court might weigh. Naming a companionship app in the psychological-harm example does not itself outlaw the category; it signals which product features, anthropomorphic cues and engagement-driving emotional dependence among them, the Commission considers relevant evidence when Article 5(1)(a) is actually applied to one.

What it leaves open

Because the guidelines are non-binding, a company can dispute the Commission's reading in court, and the document itself says so. It does not say that any named companion app violates Article 5; the passage is an illustrative composite, not a finding against a product. Editorially, the gap between the announced draft and the adopted text is worth noting whenever a source cites 4 February 2025 as the guidelines' publication date, since that date describes the draft's approval, not the final Communication.

• Does the cited version of the guidelines carry the July 2025 Communication number, or an earlier draft?
• Does the product in question rely on anthropomorphic cues to build emotional dependence, in the guidelines' own terms?
• What weight should a non-binding Commission reading carry against a company's own risk assessment?

The guidelines do not change what Article 5 prohibits; they show, through a companion-app example the Commission chose deliberately, what the Commission thinks the prohibition is for.

Sources & reading trail

Adopted text of the guidelines, including the companion-app example under Article 5(1)(a) and the non-binding status statement.

Source published: 29 July 2025 · Retrieved: 16 September 2026

Records the 4 February 2025 approval of the draft's content and a 31 July 2025 last-update marking the adopted text's arrival.

Source published: 4 February 2025 · Retrieved: 16 September 2026

Company documents, filings, studies and official records establish the record; the reading and the questions are Lovebot Journal editorial analysis. This retrospective draft does not imply the site published on the event date.

Continue reading

• The EU AI Act bans manipulation and requires bot disclosure
• A working paper found AI companions using tactics at goodbye
• Anthropic found companionship use under half a percent of chats
• Browse the complete the archive

Published September 18, 2026, not on the date of the event described.