Children are already living with AI.
It is influencing how they learn, search, communicate, create and make sense of the world. This is no longer a future scenario. UK research from Internet Matters, Me, Myself and AI (July 2025), found that 64% of children aged 9 to 17 surveyed had used an AI chatbot, with eight in ten of those users saying they use them weekly.
But children are not only using AI directly.
AI is increasingly present within the systems and services surrounding their lives. It can be used in education to evaluate learning or influence access and progression, in welfare and public services to support triage and decision-making, and across online platforms to recommend or moderate the content children see.
Some of these uses can have significant consequences. The EU AI Act, for example, classifies certain uses of AI in education as high-risk, including systems used for admissions, evaluating learning outcomes and determining the appropriate level of education a person receives.
So where is the child's voice in all of this?
When we design an AI system, develop a policy or introduce new technology that affects children, are we genuinely listening to them?
Are we considering different ages, abilities, cultures and lived experiences?
And are we involving children early enough to influence what we build, rather than asking for their views once the important decisions have already been made?
For me, this is where child rights, safeguarding and AI governance need to come together.
We cannot talk about child-centred AI if children are absent from the decisions that shape it.
Child participation is a right, not an optional extra
There is already a clear legal foundation for this.
Article 12 of the UN Convention on the Rights of the Child establishes the right of children who are capable of forming their own views to express those views freely in matters affecting them, with their views given due weight according to their age and maturity.
In the digital context, General Comment No. 25 on children's rights in relation to the digital environment makes this particularly relevant. The digital environment described by the UN Committee includes artificial intelligence, automated systems, algorithms and data analytics. Importantly, children themselves were consulted in the development of the General Comment.
This means child participation in AI governance should not be viewed simply as good engagement or an innovative design technique. It is part of taking children's rights seriously in the digital environment.
But meaningful participation requires much more than inviting children into a room and asking what they think.
Are we genuinely listening?
Good participation begins before a technology, policy or service has already been designed.
Children should have meaningful opportunities to help identify the problems technology is intended to address, explain how they experience existing systems, identify risks adults may not see, and contribute to discussions about what an acceptable solution might look like.
Professor Laura Lundy's widely used model of child participation provides a useful framework for putting Article 12 into practice. It identifies four interconnected elements: Space, Voice, Audience and Influence.
Children need a safe and inclusive space to form and express their views. They need appropriate information and support to enable their voice. Their views need an audience with the power and responsibility to listen. And critically, their participation needs the possibility of influence.
For me, that last element is particularly important.
Meaningful child participation is not about simply being able to say that children were consulted. The real question is whether their participation had influence.
If we ask children for their views only after the key decisions about an AI product or policy have already been made, are we genuinely involving them, or are we asking them to validate decisions already made by adults?
Are we listening to all children?
There is no single experience of childhood.
A six-year-old and a seventeen-year-old will experience technology differently. So will children with disabilities, neurodivergent children, children from different cultural and linguistic backgrounds, children in care, children experiencing poverty, and children with different levels of digital access and literacy.
Meaningful participation therefore needs age-appropriate and accessible methods.
For younger children, this may mean drawing, storytelling, play, visual prompts or exploring scenarios rather than relying on formal interviews or surveys. Adolescents may engage through workshops, peer discussions, digital activities and co-design sessions.
We also need to think carefully about which children are easiest for us to reach.
If we repeatedly consult confident, digitally literate children who already have access to participation opportunities, we risk designing systems around a very narrow experience of childhood.
Inclusion needs to be intentional.
Are we safeguarding the participation itself?
In my work across child rights, safeguarding and AI governance, this is one area I believe we need to pay much more attention to.
When we invite children to participate in the design of AI or in policy discussions, we also take on a responsibility to safeguard that participation.
What information are we asking children to share? Are we collecting personal experiences, sensitive information or data about harm? Do children understand how their contributions will be used? Is participation genuinely voluntary? Can they withdraw? Who has access to the information they provide?
And what happens if a child discloses abuse, exploitation, distress or another safeguarding concern during the process?
These questions need answers before participation begins, not after a problem occurs.
Participation must never expose children to additional harm in the name of making technology safer for them.
Safeguarding therefore needs to be designed into child participation from the beginning, with appropriate consent and assent processes, privacy protections, trained facilitators, clear reporting pathways and careful consideration of how children's contributions are recorded, stored and shared.
Child participation and child safeguarding should strengthen one another.
We already know this can be done
There are useful examples to build from.
As part of its AI for Children work, UNICEF consulted 245 adolescents in Brazil, Chile, South Africa, Sweden and the United States about their experiences, concerns and aspirations relating to AI. Their perspectives helped shape UNICEF's Policy Guidance on AI for Children.
That matters because it demonstrates something very practical. Children can contribute meaningfully to complex conversations about artificial intelligence when we create the right environment, provide appropriate information and use methods that enable them to participate.
UNICEF's subsequent guidance has continued to emphasise the inclusion of children in child-centred AI. The current edition, Guidance on AI and Children, Version 3.0 (December 2025), is addressed to policymakers, public organisations and businesses and sets out ten requirements for AI policies and systems that uphold child rights, ranging from regulatory oversight, safety, data protection and non-discrimination to transparency and accountability. Requirement 8, "ensure inclusion of and for children", asks organisations to involve children in the design and governance of AI rather than only to protect them from it. The evidence base for this edition included a twelve-country UNICEF study with children and caregivers, so the guidance itself was shaped by the participation it recommends.
This should not be exceptional. It should increasingly become part of how we approach the design and governance of technologies that affect children.
The question for governments, technology companies, schools, regulators and organisations should therefore move from:
"Should we consult children?"
to:
"How will children meaningfully participate throughout this process, and how will we demonstrate the influence their participation had?"
That is a much higher standard. A child rights impact assessment is one structured way to meet it, because consulting children is a step in the assessment itself and the record of what changed as a result is part of the output.
Protecting children and listening to children are not competing principles.
We need to do both.
In fact, listening to children can strengthen safeguarding. Children experience technologies differently from the adults designing, regulating, purchasing or governing them. They may identify risks, unintended uses and opportunities that are simply not visible from an adult perspective.
As AI continues to evolve, I keep returning to a very simple principle, adapted from the disability-rights movement's "Nothing about us without us":
Nothing about children without children.
Responsible AI should not only be safe for children.
Children's rights, experiences and voices should help shape it from the beginning.