Toys are covered by Directive 2009/48/EC on the safety of toys, listed in Annex I Section A. An AI system that is itself a toy product, or is intended to be used as a safety component of a toy, can be high risk under Article 6(1) where the toy must undergo third party conformity assessment under the toy legislation. The high risk requirements are then assessed through the toy conformity procedure. Products used by children also raise child safety, data protection and prohibited practice questions that must be checked separately, and this guide does not decide whether any named toy is high risk.
What toy legislation is relevant?
Directive 2009/48/EC on the safety of toys sets the safety framework for toys placed on the market in the Union, including essential safety requirements, conformity assessment and the responsibilities of manufacturers, importers and distributors. It is the instrument listed in Annex I Section A for toys. Toy safety law is detailed and product specific, covering matters such as mechanical and physical properties, flammability, chemical properties, electrical properties, hygiene and radioactivity, with age warnings and instructions where relevant. The AI Act does not replace that framework. It adds a classification route where the Article 6(1) conditions are met, and the AI requirements are then assessed within the toy conformity procedure. A toy manufacturer must therefore work with both instruments together. The toy file should cite the toy directive provisions relied on, the conformity procedure followed and the AI analysis, so that a market surveillance authority can follow the combined reasoning. A new toy safety regulation has been discussed in Union lawmaking, so the current toy instrument and any successor must be checked on the date of the decision.
When can a toy with AI be high risk?
A toy with AI can be high risk under Article 6(1) only where both cumulative conditions are met. First, the AI system must be itself a toy product covered by the toy legislation, or must be intended to be used as a safety component of such a product. Second, the toy must be required to undergo third party conformity assessment under the toy legislation. Not every toy requires third party assessment, and not every AI function in a toy is a safety component. A conversational toy that responds to a child, a toy that moves, a toy that connects to a network and a toy that recognises images or speech all present different questions. The safety component question turns on the function and on the consequence of failure, as set out in Article 3. A function that controls movement, heat, sound levels or access to network content can be safety relevant in a way that a simple sound effect is not. The record should describe each AI function separately, rather than treating the toy as a single AI system with a single answer. Where a toy contains several AI functions, some may be safety relevant and others may not, and the file should show that distinction.
For the broader regulatory framework, see products, machinery and vehicles overview.
What child specific questions sit outside Article 6(1)?
Classification under Article 6(1) is not the only child specific question. The AI Act contains prohibited practices in Article 5 that must be screened before classification, including practices that exploit vulnerabilities. Whether a particular toy feature falls within a prohibition is a legal question that needs the full Article 5 text and the facts of the feature, and this guide does not answer it for any named toy. Transparency duties in Article 50 may also apply, for example where a child interacts with an AI system or where synthetic content is generated. Data protection law applies where personal data of children is processed, with its own rules on age, consent and safeguards, and those rules are outside the AI Act. Child safety standards, content moderation, parental controls and the design of connected toys raise further questions under product safety and consumer protection law. A toy team should therefore run separate screens for prohibited practices, transparency, data protection and toy safety, and should not treat a clean result on one screen as a clean result on the others. Each screen should have its own record and its own confirmation route.
How should connected and talking toys be examined?
Connected and talking toys combine product safety, network security, data and content questions in one product. The examination should start with the data flows: what is recorded, where it is processed, how long it is kept, who can access it and what controls the adult who buys the toy can exercise. It should then examine the content flows: what the toy can say or show, how responses are generated or selected, what filters or safeguards apply, and what happens when a child asks an unexpected question or uses unexpected language. It should then examine the physical and electrical safety of the toy, including sound levels, small parts, batteries and connectivity failures. Finally, it should examine the conformity route: which procedure applies under the toy legislation, whether third party assessment is required, and how the AI requirements are addressed in that procedure. Each of those examinations needs different evidence, from data flow diagrams to test records to content test logs. A single test that the toy talks correctly on one occasion does not answer the safety, data or content questions. The file should show the tests performed, the conditions and the results, including failures and the action taken.
What should childcare products that are not toys do differently?
Childcare products that are not toys, such as baby monitors, feeding products, nursery furniture or childcare equipment, may fall under different product legislation or under the general product safety framework rather than the toy directive. The AI Act product route depends on the legislation that covers the product, so a childcare product that is not a toy cannot be assessed by copying the toy analysis. The team must first identify the legislation that covers the product, check whether it is listed in Annex I, and then apply Article 6(1) on that basis. If no Annex I legislation covers the product, the product route does not apply in the same way, but other AI Act provisions, such as prohibited practices and transparency duties, and other product safety and data protection duties, may still apply. The distinction between a toy and a childcare article is therefore a threshold question that should be recorded with reasons, including the intended age group, the presentation of the product and the way it is marketed. Marketing that presents a childcare product as a toy, or a toy as a childcare product, can create confusion and should be examined as part of the intended purpose evidence.
Continue with Annex I Section A and Section B.
What does this guide not decide?
This guide does not decide whether any named toy or childcare product is high risk, does not reproduce the toy safety requirements, and does not answer prohibited practice, data protection or content questions for a specific product. It does not state the current position on toy safety reform, which must be checked in the current Union texts. A manufacturer of a toy or childcare product with AI functions should confirm the product legislation, the conformity procedure, the Article 5 screen and the data protection position with a competent authority or legal adviser before placing the product on the market. Children are a vulnerable group for the purposes of several legal regimes, and a general guide is not a substitute for product specific review.
Frequently asked questions
Is a talking toy automatically high risk?
No. Article 6(1) requires both that the AI system is a toy product or a safety component of a toy and that third party conformity assessment is required under the toy legislation. A talking function must be examined for its safety role and for the conformity procedure that applies to the toy. Prohibited practice, transparency and data protection screens must also be run separately.
Does the toy directive still apply alongside the AI Act?
Yes. Directive 2009/48/EC remains the toy safety framework, and the AI Act high risk requirements for a toy that meets Article 6(1) are assessed through the toy conformity procedure. Both instruments must be addressed in the conformity file. Any successor toy safety legislation must be checked in its current form.
What if a product is used by children but is not a toy?
The product legislation that covers that product must be identified first. If it is not toy legislation, the Annex I analysis follows the legislation that does cover it, or the general product safety framework where no specific legislation applies. Child specific data protection, transparency and safety questions still need separate examination, whatever the product category.
Should a toy team rely on this guide for a launch decision?
No. A launch decision needs the product file, the toy conformity assessment, the Article 5 screen, the data protection assessment and confirmation from a competent authority or legal adviser. This guide explains the questions to ask and the records to keep, but it does not make the launch decision.
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 ↗