An Artificial Intelligence in Education Community of Interest Event
As artificial intelligence becomes ever more embedded in classrooms around the world, questions of safety and safeguarding are no longer theoretical; they are urgent and practical. Our latest AI in Education Community of Interest session, co-organised with the World Bank and EdTech Hub’s AI Observatory, brought together global perspectives from government, civil society, and multilateral organisations to explore what it truly means to protect learners in an AI‑enabled education system.
Rather than framing safety as a constraint on innovation, speakers emphasised it as a foundation for trust, equity, and long‑term impact. From children’s rights and data protection to evidence, governance, and the role of teachers, the discussion highlighted that safeguarding in AI is ultimately about designing systems that prioritise learners’ wellbeing, agency, and educational outcomes over extractive or purely commercial interests.
Explore insights and discussion questions from this special Community of Interest (CoI) session held on 26 March 2026 to learn more about this critical topic.
Featured Speakers
- Alice Carter – East Africa Director, Brink + Innovation Lead, EdTech Hub
- Moderator – Daniel Plaut, Global Learning Lead, EdTech Hub AI Observatory and Action Lab
- Carlos Ferrari, Innovation Manager, UNICEF
- Pedro Hartung, Executive Director, Alana Foundation
- Sam Howe, EdTech & AI Policy Lead, Commonwealth & Development Office (FCDO)
Watch the Webinar
Key takeaways
Here are four key takeaways, highlighting that effective AI in education is not inevitable, nor purely technical. Safety, rights‑based governance, and intentional design are emerging as essential to building trust, protecting learners, and ensuring AI supports, rather than reshapes, the human, relational foundations of education.
1. Safety is what makes AI tools credible and scalable
One message from the discussion was that meaningful AI adoption in education demands for far higher scrutiny than is currently the norm.
As Sam Howe argued, “we don’t see safety as the antithesis of innovation… safety is what makes AI tools credible and scalable.”
That credibility gap is stark. Carlos Ferrari shared that through UNICEF’s EdTech for Good evaluation process, only around 50 out of more than 1,000 tools reviewed met minimum quality standards, most failing on safety, privacy, or evidence of impact. Crucially, safety was framed as more than technical robustness or accuracy. It also means creating learning environments where children are protected from harm, exploitation, and unintended cognitive or emotional effects, and where teachers can engage with confidence. The takeaway was clear: without strong standards, governance, and evidence, AI risks undermining trust in education systems rather than strengthening them. Safeguarding, done well, is what separates hype from tools that genuinely add value.
2. Responsible AI requires shifting from market‑led adoption to public‑interest design
Safeguarding AI in education is ultimately a question of intent, incentives, and governance.
Pedro Hartung challenged the dominant extractive business model head‑on, arguing that “data extraction and commercial exploitation of children’s attention is the worst possible model, especially in education.”
Instead, Pedro emphasised a rights‑based approach grounded in the right to education, where tools are evaluated by whether they support children’s full development, cognitive, social, emotional, and civic. Brazil’s emerging framework for digital regulation demonstrates what this can look like in practice: strong data governance, limits on commercial influence, transparent accountability, and procurement standards that demand evidence of educational value. The discussion also highlighted that AI safety is contextual. Tools that appear ‘safe by design’ can be unsafe in practice, particularly in low‑ and middle‑income contexts lacking infrastructure or bargaining power. The clear message: responsible AI requires shifting from market‑led adoption to public‑interest design, continuous oversight, and human‑centred learning systems that technology supports—but never replaces.
3. AI in Education is a choice, not an inevitable path
The presence of AI in education does not mean every application is necessary, appropriate, or beneficial. As the moderator, Daniel Plaut noted, “we have more choice than we often say about how AI can be used in education.” The panel argued for intentional deployment, identifying where AI genuinely adds value while protecting the cultural, relational, and community‑driven aspects of education that cannot be automated. Effective safeguarding, in this sense, includes deciding where not to use AI. This reframing empowers education leaders to resist hype cycles, make context‑specific decisions, and ensure technology serves pedagogy rather than reshaping it by default.
4. Safeguarding requires both system‑level guardrails and child‑level agency
The discussion closed with a powerful reminder that safety cannot rely solely on technical controls or regulation. As Sam Howe put it, “safety is a dynamic, behavioural, ever‑evolving thing.”
Alongside governance, standards, and safeguards built into systems, children must be equipped with the skills to question, reflect, and seek help when encountering risks. Carlos Ferrari reinforced this with an analogy: we don’t send children into traffic unaided, we build roads, rules, and protections that earn their trust. Together, the speakers argued that responsible AI in education depends on systems that are safe by design and learners who are empowered, supported, and grounded in human relationships, curiosity, and critical thinking.
Further questions from the session
The following questions were posed by community members. We’re sharing to help stimulate further discussions and knowledge exchanges. Please note that some questions may have been edited for spelling or clarity.
Structural Inequality in AI Systems
- How can governments regulate AI effectively when technical expertise, data infrastructure, and enforcement capacity are limited?
- We have the same context for integrating AI as for using ICT in education: educational policies, learning programs, training, among other things. Given the equity challenges that already exist in EdTech, how can we plan for equitable AI in our low- and middle- income countries?
Technological Alternatives and Digital Sovereignty
- What about “small AI” (edge models) deploying more open-source local AI solutions instead of the big global LLMs (hosted in the west)? Could that be a solution to some of the data, privacy, and contextualization challenges?
Collaboration and system readiness
- How can we collaborate to integrate AI in education into a country that is just starting to integrate ICT into education in the most inclusive way?
- The example of Brazil hopefully will positively influence many other countries. But it took roughly 20 years to become effective (after the expansion of smartphones). How can we speed up this process to address the current challenges and risks of Ai in education?
Resources
The following resources were shared by community members and participants. These have not been reviewed by the World Bank or EdTech Hub, but are useful indicators of what conversations, evidence, and methods are being explored in the sector.
Resources from the World Bank
- Artificial Intelligence Revolution in Education: What You Need to Know
- 100 Student Voices on AI and Education
Resources from EdTech Hub
- Signal Library is Now Live: The First Ever Signals Database of Its Kind
- Mapping a Theoretical Framework for Education in the Age of AI
Resources from FCDO
- Generative AI: product safety standards
- Education – Incubator for AI
- Global AI for Learning Alliance (GAILA) – Call to Action
Resources from UNICEF
Other resources
- Meta and YouTube Found Negligent in Landmark Social Media Addiction Case
- AI + LEARNING DIFFERENCES Designing a Future with No Boundaries
This is part of an ongoing series hosted bythe World Bank and EdTech Hub’sAI Observatory and Action Lab. The AI Observatory is made possible by support from UK International Development. Please follow along and join the conversation on LinkedIn!
