Human-centered AI Literacy in Schools: Evidence-based Multi-stakeholder Recommendations for Primary & Secondary (K-12) Education

Synthesis: Governments are splitting into two camps on generative AI in schools — scaling tools or restricting them — and the iRAISE report argues that both leave the same work undone. Drawing on a structured review, policy documents from France, Canada, Egypt and Qatar, and consultations with officials and NGOs, it reframes AI literacy as the newest layer of digital and media literacy rather than a separate subject. Its central claim is an implementation gap: most countries name media and information literacy in national frameworks, far fewer put it into curricula in any depth, and almost none measure whether students acquire it.

The source is a multi-stakeholder policy report, not a peer-reviewed study: it presents no new primary data, its figures come from published research and official statistics, and its country analysis rests on document review and interviews with officials, NGOs and experts rather than with children, families or teachers.

Key Findings

1. Human relationships remain the condition learning depends on. The report argues AI should support teachers' relational, ethical and evaluative work, not stand in for it.
2. Schools have a civic function that personalized pathways can erode. Socialization happens in the hidden curriculum, and fragmenting students' routes through it weakens shared ground.
3. Effective programs rest on three pillars. Resources, content design and pedagogy all matter, and the resource most often missing is time in the timetable.
4. Frameworks are abundant; curricula and measurement lag. Most countries name media and information literacy in national frameworks, few carry it into curricula beyond technical skills, and fewer still measure it.
5. NGOs and universities are the human infrastructure. They turn national strategies into local practice and reach marginalized youth, but short funding cycles limit evaluation.
6. Coordination across ministries is a precondition. Education can coordinate, but no single ministry can govern a technology that touches health, child protection, justice and culture.
7. Human factors decide outcomes. In all four countries studied, teacher confidence, parental mediation and community trust mattered as much as policy design.

What the report is, and what it measured

The authority here is breadth, not new evidence. The report combines a structured review of global evidence, analysis of policy documents and official statistics for four countries chosen to contrast governance models, and consultations with officials, NGOs and experts in each. The systems differ by orders of magnitude: France has about 11.6 million students, Canada about 5.5 million in public schools, Egypt about 28.5 million pre-university students, and Qatar about 419,000.

The report's argument about purpose comes from Gert Biesta's three aims of schooling — qualification, socialization and subjectification. As information becomes freely available outside school, the authors argue, the second and third become more important rather than less, which is the part of the case that procurement decisions tend to miss.

Frameworks are not curricula

The sharpest distinction in the report is one the AI literacy literature usually blurs. A competence framework such as DigComp or France's CRCN states what students should be able to do; a curriculum sequences that into lessons and gives it time.

The gap between the two is measurable. UNESCO's 2025 mapping found that 171 of 194 Member States referenced media and information literacy or related competences in national frameworks, but only 84 (43%) included elements beyond technical digital skills in school curricula. An earlier count put the baseline lower still: only about half of countries worldwide had defined the digital skills they expect learners to acquire. Measurement is thinner than either, and the report treats the forthcoming PISA 2029 assessment of Media and AI Literacy as the first comparable instrument at scale.

Where evaluations exist, effects are narrower than advocates claim. Two recent US evaluations found that programs improved students' knowledge, but changes in behavior were mainly observed among older students.

Four systems, one conclusion

France pairs a strong sovereign framework — the CRCN, the Pix certification and a data-protection gateway — with territorial inequality and a teacher-training bottleneck. Canada has NGO programs of broad reach but unmeasured depth and short funding cycles. Egypt scaled infrastructure quickly but faces equity gaps, teacher anxiety about technical failure and parental distrust. Qatar closed the access divide and now confronts wellbeing concerns and a shortage of Arabic-language AI content.

The conclusion across all four is the report's most portable finding: infrastructure and content did not guarantee outcomes. Teacher confidence, parental mediation and community trust did as much work as policy design. That cuts against procurement-led strategies that treat devices and platforms as the intervention, and it supports coordination across ministries rather than within one.

What this means for practice

• Policymakers: legislate the timetable, not just the framework. Decide whether digital and AI literacy is its own subject, is integrated into existing ones or runs as cross-curricular projects, and state what it replaces.
• School and district leaders: treat teacher confidence as infrastructure. In all four countries, anxiety about tools shaped whether anything was used at all.
• Teacher educators: place training inside paid working time, beginning in initial teacher education and differentiated by subject and role rather than delivered as one generic AI course.
• Curriculum designers: build AI literacy on media and information literacy rather than beside it, which is also how the PISA 2029 assessment will treat it.
• Funders: prefer multi-year, recurrent commitments. Political cycles are shorter than the time teacher training and evaluation need to show effects.

Limitations

• Most quantitative evidence reviewed is correlational, drawn from cross-sectional studies and meta-analyses; reverse causation and confounders such as socioeconomic status are rarely fully controlled.
• Children, adolescents and parents were not consulted directly, so their perspectives enter only through officials, NGOs, experts and the survey literature.
• The four case studies include no low-income country and no Asian or South American system, so transfer to very different settings is untested.
• Country indicators are approximate and, by the authors' own note, not fully harmonized.
• The report proposes a Universal Learning & AI Literacy Toolkit but evaluates no implementation, so its recommendations are design proposals rather than tested interventions.

Connected Concepts

• AI Literacy — where the framework-to-curriculum gap lands
• Information and Media Literacy — the foundation the report builds on
• Teacher AI Competency — training, time and confidence as working conditions
• Educational AI Policy — inter-ministerial coordination and procurement

Connected Articles

• AI adoption readiness among Ukrainian education managers: Barriers, typologies, and policy implications — readiness as a management and policy problem rather than an individual skill
• Generative Artificial Intelligence Policy: A Qualitative UNESCO Framework Analysis — how UNESCO policy guidance frames national AI literacy work

Citation

Cerioli, M. (2026). Human-centered AI Literacy in Schools: Evidence-based Multi-stakeholder Recommendations for Primary & Secondary (K-12) Education. iRAISE (everyone.AI & Paris Peace Forum).