AI governance in education is not chiefly about choosing tools. It is about whether schools and systems can still act in the public interest when those tools arrive.

UNESCO’s new consultation on artificial intelligence in education is not an invitation to choose a favourite chatbot. It is a request to decide what education systems must retain control over as AI becomes embedded in platforms, assessment, administration and teaching.

The Global consultation on education in the age of AI was launched in September 2026. UNESCO is inviting experts, academics, government officials and other stakeholders to review a discussion paper and six commissioned background papers. Comments are due by 15 October 2026, and UNESCO says that the feedback will inform a compilation of policy briefs planned for Digital Learning Week 2027. (UNESCO, 2026a)

That distinction matters. The papers are not final UNESCO policy. They are background contributions intended to stimulate discussion. Their usefulness lies less in presenting a ready-made global rulebook than in making a more difficult proposition concrete: AI governance is about education’s public purpose, not simply technology adoption.

A common-good starting point

UNESCO’s foundation paper frames education as a common good and argues for deliberative governance. Put plainly, schools and colleges exist to advance learning, social relationships, inclusion, judgement and democratic life—not to serve as low-risk testing grounds for commercial products. (UNESCO, 2026b)

The consultation begins from a practical problem. AI is changing quickly while many education systems lack suitable regulation, institutional capacity and a shared understanding of risks and opportunities.

The background papers make this broad proposition more practical. Sovereignty asks who sets the terms. Age appropriateness and safety ask what learners should be exposed to and under what safeguards. Procurement and total cost of ownership ask who bears the obligations created by a tool. Explainability asks who can understand, challenge and reverse consequential decisions.

Together, the papers shift attention from “Can this platform be used?” to “Can a public education system justify, oversee and, if necessary, stop its use?”

Sovereignty is capacity, not a national chatbot

The paper on digital and AI sovereignty offers a useful corrective to simplistic calls for domestic AI. It treats sovereignty as a set of connected capacities rather than self-sufficiency in every layer. These include pedagogical and epistemic capacity, linguistic and cultural representation, data governance, infrastructure and financial control, and jurisdictional authority. (Benotti et al., 2026)

A school system can control its data yet remain dependent on foreign model updates. It can offer an AI tool in a local language while lacking the ability to test whether its outputs respect local knowledge or cultural context.

For teachers and VET practitioners, the implication is not that everyone must become an engineer. It is that AI literacy should include judging data, hallucinations, stereotypes and uncertain outputs, rather than memorising a vendor’s interface. For system leaders, sovereignty means treating interoperability, portability and a genuine right to exit as educational safeguards.

The paper also resists a familiar false choice. Capacity may be built through smaller, fit-for-purpose models, regional collaboration and local benchmarks, not only through expensive frontier infrastructure. But local capability is not cost-free. It brings maintenance, cybersecurity and updating obligations, which leads directly to the consultation’s more prosaic—and consequential—papers.

Procurement is where public purpose becomes enforceable

The procurement paper argues that purchasing is an underused lever in AI governance. Procurement can turn principles into conditions before institutional dependence hardens. A tender can require educator participation, govern data extracted from learners and staff, and refuse tools that cannot meet a public-purpose test. (Gulson et al., 2026)

It can also recognise that the decision is not complete when a licence is signed. Products change, data practices evolve and inequities may become visible only in classrooms.

The paper on total cost of ownership makes the financial companion argument. An AI system’s price is not its total cost. Continuing costs include connectivity and devices, cloud or usage charges, identity and integration, privacy, cybersecurity, technical support, evaluation, vendor management and professional learning. (Krueger & Jackson, 2026)

A “free” pilot can therefore be an expensive commitment in disguise. For under-resourced schools, especially where power and connectivity are not assured, this is not a finance-office detail. It is an equity question. A tool that works in a well-supported pilot but cannot be safely supported across a system may widen rather than reduce opportunity.

Age, safety and the case against a simple threshold

The paper on age-appropriateness refuses the neatness of a universal age cutoff. Development is variable, and chronological age is only a policy proxy. The relevant question is whether a particular AI system is demonstrably appropriate for learners of a particular age, with particular needs and abilities, in a defined educational purpose and setting. (Livingstone et al., 2026)

The authors recommend a precautionary approach. Before using AI, education systems should ask whether independent evidence supports a benefit, whether a non-AI alternative would be better, whether data processing is lawful and safe, and whether local testing covers accessibility, language, culture and disability.

This complements the paper on learner safety, which argues that safety should be a precondition of scale rather than voluntary advice issued after rollout. Schools need authority to restrict, test, expand, monitor and remove tools. (Asare, 2026)

This is especially relevant when AI arrives quietly inside a learning platform update or an assessment product, without a separate “AI procurement” moment. In VET settings, where an AI tutor, simulator or predictive system can influence progression and employability, the need for defined human oversight and a withdrawal route is hardly theoretical.

The right to an explanation is a classroom issue

The paper on explainable AI extends safety beyond harmful content. When AI supports grading, placement, admissions, behavioural prediction or personalised support, an affected learner, parent or educator needs meaningful information, a route to contest the outcome and qualified human review. (Kralj, 2026)

This is not a demand that every neural-network parameter be translated into plain language. A more realistic standard is a comprehensible reason, traceable documentation, accountable people and an effective remedy.

That is a compelling account of explainability as a procedural safeguard for the right to education. It also exposes a difficult procurement test: if a supplier cannot disclose the basis of an important recommendation before deployment, should a school be permitted to rely on it? Teachers should not be left to explain or override a system they cannot inspect while learners carry the consequences.

What the series gets right—and what remains unresolved

The chief strength of the consultation is its coherence. The papers do not mistake a technical checklist for governance. They connect learner rights, teacher professional judgement, cultural and linguistic relevance, fiscal durability and institutional authority across the whole lifecycle of AI use.

They also resist both techno-determinism and a blanket ban. An AI use can be authorised conditionally, reviewed against evidence and removed when it fails to serve its educational purpose.

The limitations are substantial, and the papers acknowledge many of them. Evidence on children, equity, social-emotional effects and Global Majority contexts remains thin. Much available research comes from better-resourced settings. Local evaluation, independent audits, rights-impact assessment, cybersecurity and meaningful human review all require money and skilled people.

The papers say less about who supplies those capabilities for small schools, rural systems and underfunded VET providers—or how educators’ time for participation and review will be protected. Opacity in advanced systems may also make a meaningful explanation difficult rather than merely neglected.

These are precisely the questions that a consultation should address. Practitioners can contribute evidence of the tools already embedded in their environments, the costs that vendors omit, the decisions learners cannot challenge and the safeguards that actually work.

The test for UNESCO’s eventual policy briefs should be demanding: do they give education systems enough agency to say not only “yes” to a useful tool, but also “not yet”, “only under these conditions” and “no” when education’s public purpose is at stake?

References

Asare, I. N. (2026). Safe enough to scale? A decision system for learner safety [Background paper]. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000399345_eng

Benotti, L., Busaniche, B., Gómez, M. J., & Trovato, M. (2026). Digital and AI sovereignty: Four evidence-informed cases [Background paper]. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000399346_eng

Gulson, K. N., Huang, X., Bello Villarino, J.-M., & Vlies, P. (2026). AI procurement in education: The missing lever in AI governance [Background paper]. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000399344_eng

Kralj, L. (2026). The right to an explanation: Towards explainable AI in education [Background paper]. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000399347_eng

Krueger, K., & Jackson, F. (2026). Beyond the price tag: Rethinking total cost of ownership for AI in education systems [Background paper]. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000399348_eng

Livingstone, S., Sylwander, K. R., El Gemayel, S., & Atabey, A. (2026). Is there a right age for AI in education? Deciding whether AI is age-appropriate and why [Background paper]. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000399343_eng

UNESCO. (2026a). Global consultation on education in the age of AI. https://www.unesco.org/en/digital-education/artificial-intelligence/consultation

UNESCO. (2026b). Sustaining education as a common good in the age of AI: The case for deliberative governance. https://unesdoc.unesco.org/ark:/48223/pf0000399349_eng

The papers discussed here are commissioned background papers for consultation and should not be treated as final UNESCO policy. The article is an analytical discussion, not legal advice.