As AI enters childhood, safety rules struggle to keep pace: WEF

Artificial intelligence (AI) is increasingly becoming part of everyday childhood, powering homework tools, games, social media feeds and conversational assistants. While the technology can support education, creativity, accessibility and skill development, it can also amplify existing harms and create new risks through personalisation, persuasion, data retention and autonomous action.

A new child-safety framework argues that protecting children in AI environments will require more than blanket bans or age restrictions. Governments, technology companies, educators, and parents need safeguards that are proportionate to the nature and risks of individual AI systems.

AI governance remains fragmented

AI regulations often do not address children in sufficient depth, while online safety laws vary significantly in how clearly they apply to AI services. International human rights law provides an important foundation, but does not offer a complete AI-specific operating model.

Differences in definitions, safeguards and enforcement capabilities can create regulatory gaps and uncertainty. Many countries also face limitations in regulatory capacity, child-protection systems, digital infrastructure and technical expertise.

The framework says there is no one-size-fits-all approach. Safeguards should be tailored to the system and the risks it presents, while avoiding restrictions that unnecessarily limit beneficial uses of AI or undermine children’s autonomy.

Access must be meaningful and equitable

Child well-being extends beyond protection from harm to include empowerment, development, agency, inclusion, privacy, dignity, beneficial opportunities and human connection.

Meaningful access to AI requires more than simply making services available. Children also need connectivity, language support, digital and media literacy, and the ability of families and schools to configure and use AI systems effectively.

Restrictions should therefore be proportionate to the nature and level of risk. Blanket restrictions could limit children's opportunities to develop digital skills without necessarily addressing the experiences most clearly linked to harm.

AI systems should also account for children with disabilities, those who speak less widely supported languages and those living in underserved communities.

Children need AI literacy

As AI becomes more embedded in education, civic life and future workplaces, children should be able to use it to ask questions, practise skills, create, experiment, access information and receive feedback.

However, exposure alone does not build capability. Children need to understand how AI works, why its outputs can be wrong, where information comes from and when human judgement is required. They should also learn to recognise automated persuasion, distinguish assistance from substitution and identify when personalisation may narrow their choices.

Parental controls are not enough

AI systems used by children should have appropriate safeguards and parental controls, but children should also have meaningful control over their experiences, depending on age, capacity and risk.

This could include the ability to pause or disable AI features, select specific content or interaction settings, decline the reuse of conversation data, request explanations and manage what they wish to see or avoid. However, the framework stresses that agency should not mean transferring responsibility to children. Companies must address risks they can reasonably control, and parental controls should not substitute for safe product design.

Privacy and human connection matter

AI systems can process highly sensitive information about children, including conversations, locations, social relationships, creative work and inferred emotions. Protecting children from harm should not require them to surrender their right to privacy.

Systems should minimise data collection and provide meaningful user controls. Children should also be protected from AI applications that imitate, alter or expose their identity in ways that undermine dignity, reputation or autonomy.

At the same time, AI should complement rather than replace relationships with parents, caregivers, teachers and peers. These relationships provide empathy, accountability, culture and real-world experiences that are central to development.

AI can nevertheless offer accessibility, reassurance and a way to ask for help, particularly for children who are isolated, disabled or worried about judgement. The key consideration is whether AI helps children reconnect with people and activities, rather than encouraging dependence on the system itself.

Age gates cannot make AI safe

The framework cautions against treating age restrictions as a universal solution. A young child, an early adolescent and a 17-year-old can have very different needs and capacities. Even children of the same age can differ in maturity, experience, ability and family support.

Age assurance can help apply differentiated safeguards, but it cannot determine developmental readiness or replace safer design, moderation, digital literacy and effective remedies.

Generative and agentic AI also introduce new challenges because these systems can maintain continuous conversations, remember interactions, personalise responses and, in some cases, act independently.

Reporting needs a stronger remedy

A reporting button is not enough when something goes wrong. Children may need clear explanations, safe choices, timely human review and access to support.

Automated reporting can be inadequate when a child discloses abuse, self-harm, sexual concerns or another vulnerability. Children may fear punishment, losing device access or automatic disclosure to parents or schools.

The framework argues that responsibility for remedies should remain with the institution controlling the relevant decision, rather than leaving children and families to navigate a complex chain of AI providers, platforms and distributors.

Ultimately, no single age threshold, parental setting or technical filter can deliver child-safe AI. The framework calls for coordinated action across developers, technology companies, governments, regulators, educators, parents and civil society, with safeguards that evolve alongside the technology and emerging evidence.