Child Centric AI

Artificial intelligence is already shaping childhood. It influences what children learn, what they see, who they interact with and how they understand themselves.

The question is no longer whether children will encounter AI. The question is whether AI will be designed with children in mind or whether children will once again be left to adapt to systems built for adults.

This page explains what child centric AI means, why it matters now, and what organisations can do immediately.

Why child centric AI matters now

History shows a repeated pattern. Powerful technologies are introduced. Evidence of harm appears early. Action comes late.

Seatbelts existed decades before they were mandatory. Child labour was widely documented long before it was restricted. In both cases, people paid the price while society delayed. AI is moving faster than any of those technologies ever did.

Generative AI reached one hundred million users in two months. AI tools are already widely used by teenagers. Most parents and schools lack clear frameworks. There is no comprehensive child specific AI regulation anywhere in the world.

We do not have decades to respond. We may not have years.

What child centric AI is and what it is not

Child centric AI is not just child safe AI. Safety matters, but safety is the minimum requirement, not the goal. Most current approaches focus on harm prevention. Content filters. Age checks. Parental controls. These are necessary, but on their own they treat children as an afterthought.

Child centric AI starts from a different question.

How do we design AI systems that actively support children’s rights, development, dignity and wellbeing by default.

The five principles of child centric AI

Child centric AI is grounded in five principles. These are practical requirements for design, governance and deployment.

Principle one. Children are not small adults. Developmental appropriateness. Children’s cognitive and emotional capacities change rapidly as they grow. An AI system that works for an adult does not work safely or appropriately for a child. Child centric AI recognises different developmental stages. It adapts tone, authority and interaction accordingly. It avoids presenting information with unqualified confidence to young users. Age appropriate interaction is not a feature added later. It is a design philosophy.

Principle two. Privacy is a right, not a preference. Privacy by default. Children cannot meaningfully consent to extensive data collection. Child centric AI minimises data collection, limits retention, avoids emotional or behavioural tracking for engagement optimisation, and gives parents and children real visibility and control. Children’s data must never be used to optimise engagement at the expense of wellbeing.

Principle three. Transparency must be real, not performed. Genuine transparency. Lengthy terms of service are not transparency, especially for children. Child centric AI clearly explains when a child is interacting with an AI system. It explains what the system is doing in age appropriate language and at the moment it matters. It gives children a genuine ability to say no without losing access.

Principle four. Fairness and inclusion by design. Inclusive AI. AI systems reflect their training data. Children from marginalised communities are consistently underrepresented. Child centric AI deliberately includes children from diverse linguistic, cultural and socioeconomic contexts in design and testing. It works in minority languages. It functions in low bandwidth and offline environments. If AI only works well for children in wealthy, connected contexts, it is not child centric.

Principle five. Agency and empowerment. Active empowerment. Child centric AI should actively support children’s agency. It should help children learn how to think, not just what to think. It should support creativity rather than replace it. It should encourage healthy independence rather than emotional dependency. Systems that optimise for engagement at all costs fail this principle.

Three lenses for action

Whether an organisation is building AI, buying AI or governing AI, child centricity shows up in three places.

1. Design. Is this built with children in mind. Child centric design requires assessing child impact before launch, not after harm occurs. A basic child impact check asks four questions. Could a child encounter this system. What happens if they do. What data is collected about them. Who is accountable if something goes wrong. If these questions cannot be answered clearly, the system is not ready.

2. Governance. Does the organisation have a child specific framework. Most AI ethics frameworks mention vulnerable users. Few mention children explicitly. Child centric governance requires structured child impact assessments, named accountability with authority to stop deployment, and ongoing monitoring after launch. A system that is safe today may not be safe six months later.

3. Collaboration. Who else needs to be in the room. Child centric AI cannot be built in isolation. It requires child development expertise in product decisions, testing in real world contexts rather than only labs, and open sharing of lessons learned. Pre competitive collaboration on child safety is not a risk. It is a responsibility.

Child centric AI must work for children in low connectivity environments, for children who speak minority languages, and for children in humanitarian and climate vulnerable contexts

The three stages of organisational maturity

Most organisations fall into one of three stages.

Stage one. Awareness. The organisation recognises that children might use its AI systems.

Stage two. Mitigation. Safeguards such as filters or age checks have been added.

Stage three. Child centricity. Children are explicitly considered in design, governance and accountability from the start.

The gap between stage one and stage three is not technical. It is organisational will.

Four things you can do this quarter

Action does not require a new budget or new regulation. This quarter, organisations can take four concrete steps.

1. Audit existing AI products and services for child exposure.
2. Ask whether a child could encounter them. Update AI governance frameworks to explicitly include children, naming risks, responsibilities and safeguards.
3. Run one child impact assessment on the highest risk product or service.
4. Invite a child development expert into one product or deployment review and listen to what they notice.

Small actions taken now prevent large harms later.

Designing for the children who need it most

Child centric AI must work for children in low connectivity environments, for children who speak minority languages, and for children in humanitarian and climate vulnerable contexts.

If AI works for a child in a rural school with unreliable internet, it will work everywhere. Designing for the margins creates systems that are stronger, fairer and more resilient for all children.

The seatbelt moment

Seatbelts did not slow innovation. They made cars safer by default and improved trust for everyone. Child centric AI is the same.

It is not a brake on progress. It is a foundation for trust. It is not a constraint on innovation. It is a design principle that makes AI better. The seatbelt moment is now.

The question is not whether we can act, but whether we will.

Read more about UNICEF’s toolkit that helps businesses conduct child rights impact assessments in relation to the digital environment: Assessing child rights impacts in relation to the digital environment | UNICEF Child Rights and Business