AI is reshaping childhood. How do we protect the next generation?

As artificial intelligence (AI) becomes part of how children learn, communicate, create and play, governments face a difficult responsibility in how to protect children from harms while also preparing them to participate safely and meaningfully in an increasingly AI-mediated world. Governments and industry actors have a duty to protect children, including online. The UN Committee on the Rights of the Child affirms in its landmark General Comment No. 25 that children’s best interests must be a primary consideration in the regulation, design and use of digital technologies.

Online safety policy covers a wide range of digital harms, including bullying, exploitation, harmful content, privacy violations, addictive design, manipulation and risks to mental health. These harms are serious, but they do not necessarily have the same causes or require the same responses.

Children’s rights and well-being provide the necessary starting point. Safety matters, but so do privacy and dignity, agency, learning and development, human connection, inclusion and access. Protecting children therefore requires considering not only whether they can access a technology, but what happens when they do.

Recent policies increasingly recognize this approach. Australia’s social media age restrictions require certain platforms to prevent under-16s from holding accounts. Australia’s eSafety Commissioner does, it must be noted, explicitly describe the restriction as one part of a holistic regulatory approach that also includes safety-by-design, digital literacy, education, complaints mechanisms and industry accountability.

The European Commission’s proposed EU KIDS Act similarly combines access and design. Children under 13 would not be allowed to have a social media account, while 13- and 14-year-olds could use parent-managed accounts with limited functionality and a daily limit of one hour. Autonomous accounts would begin at 15. The proposal also addresses recommender systems, unwanted contact, addictive design and privacy, while extending protections to online games, AI companions and chatbots.

But access restrictions cannot make the underlying technology safe. Risks from harmful design, manipulation, exploitation or inappropriate AI interactions do not suddenly disappear when someone turns 15, 16 or even 18. This is why access restrictions and safety-by-design combined in regulatory frameworks.

As access rules expand, however, policy-makers also need to consider their effects across all dimensions of children’s well-being. For example, should an hour spent messaging friends, learning about current events or creating something be treated in the same way as an hour of scrolling on social media?

Digital environments provide benefits alongside risks. Access to information is itself an important right. The Reuters Institute’s 2026 Digital News Report found that 52% of 18–24-year-olds globally now identify social media, video networks and AI chatbots as their main way of getting news. Restricting access to children may reduce exposure to certain harms, but policy-makers also need to consider effects on information access, participation, connection and development of media and AI literacy.

Questions of proportionality have already emerged. France’s Constitutional Council previously blocked legislation requiring parental authorization for children under 15 to access social networks, including because of concerns about freedom of expression and how age verification affects privacy.

Family circumstances also differ. Parental controls can be valuable, but parents vary in digital literacy, available time and resources. Children may live across multiple households, with grandparents or other guardians, or lack an engaged or digitally capable adult. Regulation should avoid assuming that every child has the same family environment or support.

AI further complicates the distinction between access and design.

New York City’s recent one-year moratorium on student-facing generative AI through eighth grade (roughly ages 13–14) provides another example of an access-based intervention that has merit – much like restricting calculators so students first learn core mathematical skills.

However, children will encounter AI outside classrooms. Increasingly, they may not even choose to access AI as a distinct product. AI is being integrated into search, games, operating systems, education and communication, while AI agents will increasingly act across several of these environments.

This makes some regulatory boundaries difficult to define. The proposed KIDS Act seeks to address AI companions and chatbots that simulate interpersonal relationships or create emotional dependency. The concern is important, particularly as children develop social skills and conversational systems become increasingly personalized.

But what constitutes an AI relationship? When does repeated interaction become dependency? How should an educational tutor that remembers a child’s learning needs differ from a companion that remembers personal conversations? And what happens when a single AI agent helps a child learn, communicate, search and play?

Rather than relying only on product categories, regulation will increasingly need to consider the functionality, context and risks of AI systems.

A new World Economic Forum paper, Children’s Safety and Well-Being in the Age of AI: Best Practices and Incentives, developed through the Global Coalition for Digital Safety, frames the objective around children’s broader well-being and provides best practices for organizations across the AI lifecycle.

Regulations are needed, and access restrictions are appropriate where evidence shows that delaying access protects children during critical stages of development. However, these policies must also be continually evaluated to assess whether they deliver intended outcomes and what trade-offs they create. Safety-by-design is necessary to make the technologies they eventually encounter safer. Education and AI literacy are also needed to equip children to navigate risks that regulation cannot eliminate.

The objective should therefore not simply be keeping children away from risk until they reach a particular birthday. The digital harms affecting children also affect adults and thus will not disappear.

The challenge is to determine when access should be restricted, how design should change and how both can contribute to children’s overall well-being. As AI becomes embedded throughout the digital environment, protecting children must also mean preparing them to exercise agency, protect their privacy, maintain human connection, access information and navigate an AI-mediated digital world safely once those restrictions are lifted.