The EU AI Act in Children's Communication: What Applies Since August 2026

Law & AI, August 9, 2026, 10 min read

Since 2 August 2026 the transparency obligations of the EU AI Act have been applicable. In many marketing departments the date was treated as an IT topic. For brands addressing children, parents and families, it is not. The strictest rule for this audience has in fact applied since February 2025, yet it is rarely mentioned. This article sets out what applies concretely, what else follows from the DSA and the German JMStV, and what parents actually expect. This is not legal advice but an assessment from marketing practice.

What applies since 2 August 2026

Article 50 of the EU AI Act governs transparency. Providers must design AI systems so that people are clearly informed when they interact directly with an AI. AI-generated or manipulated content must be marked in a machine-readable way. Deployers, and in marketing these are usually the brands themselves, face disclosure obligations for deepfakes, emotion recognition, biometric categorisation and for texts on matters of public interest published without human editorial control.

Anyone following the European Commission's Code of Practice on Transparency of AI-generated Content is considered compliant. Anyone who does not must demonstrate equally suitable means. For systems placed on the market before 2 August 2026, a grace period for technical marking runs until 2 December 2026.

Important for planning: the Digital Omnibus, on which a political agreement was reached in early May 2026, postponed several deadlines. The high-risk obligations under Annex III move to December 2027, Annex I to August 2028. Article 50 was not postponed. Anyone waiting for an extension is waiting in vain.

Article 5 is the stricter rule for children's brands

The public debate revolves around labelling. For brands with a young audience, a different provision is more relevant, and it has applied since 2 February 2025.

Article 5(1)(b) prohibits AI practices that exploit vulnerabilities due to age, disability or a specific social or economic situation, where behaviour is distorted in a way that causes significant harm. Point (a) in parallel prohibits subliminal, purposefully manipulative or deceptive techniques.

This is not a marginal provision. It is a prohibition, not a transparency requirement. It applies regardless of whether a system is classified as high risk. And it hits exactly the mechanics that have been common in children's marketing for years, as soon as AI personalises and optimises them: adaptive reward logics, dynamic pricing, personalised urgency, algorithmically optimised gamification.

The practical question is therefore no longer whether a piece of content is labelled. It is whether a system systematically exploits a child's inexperience. That assessment comes before any labelling discussion.

What else applies: DSA guidelines and the JMStV

The AI Act does not stand alone. Two further sets of rules apply to family brands in 2026 and are frequently overlooked.

The European Commission's guidelines on the protection of minors under the Digital Services Act of 14 July 2025 are more concrete than the AI Act. They cover all online platforms accessible to minors, with the exception of micro and small enterprises.

In Germany, the new Jugendmedienschutz-Staatsvertrag has applied since 1 December 2025. It explicitly extends the protected interest to psychological integrity and age-appropriate development, and obliges operating system providers to offer a one-button option for youth protection settings at device level.

• Age verification must be accurate, reliable, robust, non-intrusive and non-discriminatory.
• Advertising must not exploit the commercial inexperience of children. Certain virtual currencies and loot boxes are treated as a manipulative practice.
• AI chatbots integrated into platforms need safeguards against persuasive design.
• Engagement mechanics such as streaks, autoplay and push notifications must be off by default for minors.
• Recommender systems should prioritise explicit user signals over behavioural tracking.

What parents actually expect

Regulation describes the lower limit. Brand relevance is built above it. The data situation in Germany is clear here.

The Ipsos KI-Monitor 2026 shows that 75 percent of Germans demand that products and services using AI disclose this. 52 percent would trust the answers of a generative AI tool less if advertisers influenced them. Enthusiasm for AI products has fallen from 45 to 37 percent within one year.

A 2025 study by ARD MEDIA adds to the picture: 70 percent have already noticed AI-generated advertising, 63 percent can barely distinguish it from conventional advertising. 77 percent consider labelling important, and 60 percent say transparency about the use of AI would increase their trust in a brand.

One finding from Hochschule Macromedia Hamburg together with Grabarz und Partner from July 2026, with 711 participants, is notable: AI labelling works in a context-dependent way. Labelled AI text reduces trust more than labelled AI images. Premium brands react more sensitively than FMCG brands. And younger audiences judge the use of AI in visual media more sceptically than older ones, not more tolerantly.

For family brands this means: labelling is not a risk to be minimised. It is a trust signal to be designed. Anyone hiding it loses twice, legally and with the audience.

What brands should do now

A sensible start is not a legal opinion but a stocktaking along your own processes. Four steps have proven useful.

In our experience the effort amounts to a few days. The benefit is twofold: you know your risks, and you can actively use transparency as a brand statement instead of treating it as a mandatory disclosure.

• Inventory: where in marketing is generative AI used today, in which channels, with which tools, with what human control?
• Article 5 review: which of these applications affect minors, and do they optimise for behavioural change? This review precedes the labelling question.
• Labelling concept: where, how and in which tone is the use of AI disclosed? Visible for people, machine-readable for systems.
• Approval process: who decides which AI-generated content is served to a young audience? This responsibility must be named, not implied.

Conclusion

2 August 2026 was not a cut-off date on which something new began. It was the point at which a development became applicable that has been running for children's and family brands since February 2025. Anyone who knows Article 5 has already understood the harder part.

The real task is not legal but creative: to use AI in a way that respects a sensitive audience, and to make that visible. Brands that manage this gain trust exactly where others are currently losing it.