Most conversations about AI in schools pick one lens — either "is it safe for children" or "who controls this technology" — and stick to it. A keynote at UNESCO Digital Learning Week 2026 from Mila, the Quebec AI Institute, deliberately put both in the same sentence: AI safety AND sovereignty in education. That pairing is worth unpacking, because the two ideas reinforce each other in a way that matters for any parent choosing an AI tool right now.

Who gave this keynote, and why it's notable

Mila is one of the world's leading AI research institutes, based in Montreal and closely associated with Turing Award winner Yoshua Bengio. It is not an education body — it's a frontier AI research institute, the kind of organisation more usually associated with publishing papers on model architecture than speaking directly to school policy.

That its education lead, Shingai Manjengwa, delivered a keynote at UNESCO's flagship education-and-technology event, under the title "AI safety and sovereignty in education," is itself a signal worth noting. It means the people building and studying the most advanced AI systems now see K-12 education as a domain where their expertise is directly relevant — not just a downstream market for whatever consumer AI happens to produce.

Two different problems, one keynote

Safety is the more familiar half. In an education context, it means protecting a child from harm the AI system itself causes — a wrong answer delivered with total confidence, a design pattern that manipulates rather than teaches, content that's inappropriate for a child's age, or data handled carelessly. This is the half of the conversation most parents already instinctively worry about, and it's the half UNESCO's own ministerial statement addresses directly (see learner rights and AI safety: UNESCO's principles).

Sovereignty is the less familiar half, and arguably the more structurally important one. It asks: who actually controls the AI system your child (or your country's schools) depend on? Is it auditable? Can it be switched to a different provider if something goes wrong? Or is a country's — or a family's — entire education-AI infrastructure dependent on decisions made inside a small number of foreign technology companies, with no meaningful local accountability?

The two are connected in a way that's easy to miss. A system can be "safe" in the narrow sense — it doesn't say anything harmful today — while still being unsafe in the sovereignty sense, because nobody outside the company that built it can verify that will remain true tomorrow, or knows what happens to the data it collects, or has any recourse if the company changes its product, its pricing, or its ethics. We go deeper on the sovereignty half specifically in digital sovereignty in schools, explained.

Why a research institute cares about this, not just governments

It would be easy to assume "sovereignty" is purely a government-level concern — trade policy, national infrastructure, the kind of thing that shows up in a ministerial statement rather than a parent's evening. But Mila's framing pushes it down a level: sovereignty isn't only about which country's servers your data sits on. It's about whether ANYONE — a government, a school, or a parent — can meaningfully inspect, question, or walk away from an AI system they depend on.

That's a genuinely useful frame for a parent evaluating an AI tutoring product, stripped of the geopolitics: Can you delete your child's data on request? Is there a real answer to "what happens if this company changes its approach"? Is the company's reasoning for how the AI behaves — never giving away a final answer, say, or always de-escalating to a human when something's wrong — something you can actually understand, or is it a black box you're asked to trust?

What this means for a UK parent right now

You're not choosing between "safe" and "sovereign" AI systems in some abstract policy sense — you're making a much smaller, concrete choice about which product your child uses tonight. But the Mila framing gives you two separate questions to ask of any AI tool, not one:

  • Safety: Does this product protect my child directly — from wrong information stated with false confidence, from manipulative design, from content that isn't age-appropriate?
  • Sovereignty: Do I actually understand and control what happens to my child's data and interactions, or am I trusting a black box because there's no alternative?

Both questions belong on the same checklist. We've turned the safety half into a practical list in is AI tutoring safe? A UNESCO-principles checklist — worth reading alongside this one.

FAQ

What is Mila, and why did it speak at a UNESCO education event?

Mila is the Quebec AI Institute, one of the world's leading AI research institutes, based in Montreal. Its education lead, Shingai Manjengwa, gave a keynote at UNESCO Digital Learning Week 2026 on AI safety and sovereignty in education — a sign that frontier AI research institutes, not just education ministries, now see themselves as stakeholders in how AI reaches classrooms.

What does "AI safety" mean in an education context?

In education, AI safety generally means protecting learners from harm caused directly by the AI system itself — wrong information delivered with false confidence, manipulative or addictive design, inappropriate content, or data misuse — rather than the broader existential-risk sense the term sometimes carries in AI research circles.

What does "AI sovereignty" mean for a school or a country?

AI sovereignty means a country, institution, or even a family retains meaningful control over the AI systems it depends on — which company built it, where the data goes, whether it can be audited or switched, rather than depending entirely on a small number of foreign Big Tech platforms with no local accountability.

Duke Harewood built aitutors.me for his own KS3 daughter. The sovereignty question — can I actually see and control what happens to my child's data — mattered to us before we had a word for it. Published 11 September 2026.