From principles to practice

Participants spoke about internal and external challenges associated with misalignment on AI governance approaches. This included a lack of perceived policy coherence across jurisdictions and disconnects between internal functions that must cooperate on AI governance, for example human rights, trust and safety, engineering, product, legal and compliance teams. Participants also challenged the misconception that AI governance sits with one person or one centralized function inside a company. In practice, AI governance was described as highly distributed: it happens “in every corner of the company,” varying by product and feature. This underlines the importance of prioritizing efforts to build bridges and improve alignment of approaches. 

A recurring theme was that industry actors may understand the importance of child rights and AI governance in principle, but struggle with what this means operationally. Product managers are described as overwhelmed and multidisciplinary, while engineers are focused on functionality. Participants repeatedly called for practical tools, case studies, implementation frameworks, checklists and clearer standards that can be translated into product development, governance and disclosure practices. They identified the need for resources to help teams understand where child rights fit across the product lifecycle and how to convert rights into concrete functionality and design choices. 

Participants observed that current regulatory, governance and company approaches often focus narrowly on child safety rather than the full spectrum of child rights. This was seen as limiting, particularly where safety-led approaches such as bans or trust and safety processes do not fully account for participation, access to information and other child rights. A more holistic child rights lens was repeatedly identified as necessary.

Participants highlighted that many AI tools are not designed or used exclusively by children, but operate in mixed-use environments, including shared household devices and general-purpose products intended for broad audiences. This creates a design and governance challenge: stronger child guardrails may be needed to protect children, but those same protections may reduce functionality, effectiveness or desirability for adult users. Companies need approaches that can adapt to shared devices, household use, and products with both child and adult users. This points to the importance of rights-respecting age assurance, gated experiences and minimum baseline protections across products used by children, even where children are not the only intended users.

Participants recognized the importance of children’s voices in AI design and governance, but also raised serious questions about how to do this safely, meaningfully and ethically. Uncertainty around whether children should be involved in testing, when they should be consulted, how they could participate in governance, what safeguards are required, and whether alternatives such as caregiver engagement or synthetic data could reduce risks. 

For UNICEF guidance on best practices for stakeholder engagement in child rights impact assessments in relation to the digital environment (D-CRIA), see here.

Some participants reflected that pre-deployment testing for child safety, security and non-discrimination is not yet sufficiently incentivized. Participants noted that testing takes time and resources, can slow time to market, and may create legal or reputational exposure. A credible label, certification or recognition mechanism was suggested as one way to incentivize better child safety practices before products are launched.  

Participants repeatedly identified disclosure as both necessary and difficult. Companies may lack evidence or understanding of child rights impacts, and the regulatory environment may discourage candid voluntary disclosure. Concerns included legal liability, public reaction, regulatory scrutiny, inconsistent reporting requirements and a tendency for compliance narratives to crowd out proactive transparency. One participant suggested exploring trusted, proportionate and possibly staged forms of disclosure.

UNICEF has explored the challenges and opportunities associated with robust reporting on child rights in relation to the digital environment – findings and specific disclosure recommendations can be found here.

Age assurance emerged as a prerequisite for many child protection measures, but participants also voiced many challenges in implementation. Participants noted that current mechanisms raise privacy, fairness, effectiveness and jurisdictional coherence concerns. Biometrics, credit cards and data collection were specifically identified as problematic, and participants questioned who should lead in developing rights-respecting solutions.

Participants pointed to the absence of a globally recognized baseline for child privacy and safety as a barrier to consistent implementation. They voiced interest in minimum standards, universal child safety expectations, rating systems, labels and child-specific adaptations of existing global standards. At the same time, participants acknowledged that global operationalization is difficult because norms, laws and definitions vary across countries.

Participants repeatedly returned to the problem of definitions: what counts as harmful content for children, how harm should be understood across jurisdictions, and how to operationalize a global approach without conflicting with local norms or rights such as freedom of expression. One participant queried the definition of a child (e.g. under 10s), reflecting a foundational need to establish broad understanding of children as under 18s as grounded in the Convention on the Rights of the Child. This definitional uncertainty also affects measures such as content moderation and access restrictions.

Participants repeatedly emphasized that companies and policymakers lack robust evidence on how AI affects children’s development, cognition and relationships. Specific concerns included cognitive offloading and impacts of parasocial relationships with synthetic systems. This evidence gap affects impact assessment, governance, policy design and the ability to define harms.

AI literacy was identified as a major opportunity. Participants suggested that current literacy efforts may be too focused on responsible use of today’s rapidly changing systems, which is difficult to keep current. Instead, they suggested education should include more durable foundations, such as how AI models work, while also engaging young people on what they want to learn. Media literacy curricula, explainability and transparency were connected to this need.

Participants identified time, resources, expertise, reporting burdens, product launch pressures and competing stakeholder demands as major barriers to scaled implementation. Existing Environment and Social Governance (ESG), reporting and compliance processes already create a heavy burden, and robust child rights due diligence requires sufficient internal resourcing, prioritization and executive sponsorship.