UNESCO's Digital Learning Week 2026, held in Paris under the theme "Education in the age of AI: Facts | Frictions | Frontiers," put a phrase in wide circulation that's worth taking seriously: public-interest AI. This is our own honest attempt to hold our product up against it — including the places the comparison breaks down.

We're the company behind aitutors.me, so read what follows knowing we have an obvious interest in the answer. We've tried to write it the way we'd want a competitor's version read: specific claims you can check, not adjectives.

What "public-interest AI" means

UNESCO didn't hand out a certification with that phrase — it's a framing, not a checklist. But the shape of it is consistent across Digital Learning Week's theme and the 9 September ministerial statement (signed by 25+ education ministers): AI entering education should be built with public accountability, transparency, and safety in mind — serving learners and the public good — rather than optimised purely for engagement, data extraction, or commercial advantage that runs against a child's interests. It's a response to a real pattern: plenty of "edtech" over the last decade has optimised for time-on-app or subscription renewal in ways that don't obviously serve the learner.

Where we think we align

Four specific, checkable design choices, not vibes:

The hint ladder never reaches the final answer. Our tutors are built around a Socratic method — Professor Pi, our maths tutor, and every other subject tutor, work through a structured hint ladder that gets progressively more concrete but stops one step short of just stating the answer to the problem in front of the child. The top rung is a fully worked similar problem, then "now redo yours" — never the child's own answer, handed over. That's a deliberate, hard-coded product rule, not a suggestion to the model — you can read about the method itself in our show-your-working explainer.

Conversation logs delete on a 30-day rolling window. We don't keep an indefinite record of what your child said to a tutor. Conversations age out automatically.

No third-party model training on conversation data. What your child says to a tutor isn't used to train someone else's model. That's stated explicitly in our privacy policy and terms of service, not left implicit.

A hard-fail safeguarding rule. If Mentor — our wellbeing-first guide — detects any indicator that a child may be struggling in a way beyond what a tutoring session should handle, it hard-fails to a fixed message pointing to a parent or Childline (0800 1111), every time, regardless of what the conversation was otherwise about.

Each of those maps reasonably well onto what "public accountability" and "safety" are meant to guard against: a system that quietly optimises against the learner's interest, or that keeps information from the people responsible for a child.

Where the comparison genuinely breaks down

We're a commercial product. UNESCO's framework is aimed squarely at governments and public education systems — the ministerial statement was signed by ministers, deciding what national AI-in-education policy should require. We're not a national policy. We're a company you pay a subscription to, and that fact alone means the framework doesn't transfer wholesale, however well our individual design choices happen to line up with its spirit.

Two honest limits worth naming:

  • "Public accountability" for a government means something a company can't fully replicate — regulatory oversight, public reporting requirements, democratic accountability. A privacy policy and a stated design rule are real commitments, but they're not the same category of accountability a ministry answers to.
  • We chose these design rules ourselves. No external body audited them against UNESCO's framework before this article was written. That's exactly why the honest move isn't "we're public-interest AI, trust us" — it's telling you what the specific rules actually are and letting you judge them, which is what the paragraphs above tried to do.

The actual takeaway

UNESCO's language gives you something more useful than a label to look for: a set of questions. Does the tool ever go beyond a hint to a straight answer? What happens to the data? Does anyone else train on it? What does it do when something's genuinely wrong? Those questions apply to us exactly as much as to any other AI education product a parent might be weighing up — which is the whole idea behind the checklist we've written next.

FAQ

Did UNESCO endorse or review aitutors.me?

No. UNESCO has not reviewed, mentioned, or endorsed aitutors.me. This article is our own honest reflection on where our product design happens to align with the public-interest AI principles UNESCO discussed at Digital Learning Week 2026 — not a claim of official recognition.

What is "public-interest AI" in UNESCO's framing?

It's the idea that AI entering education should serve learners and the public good — with accountability, transparency and safety built in — rather than being optimised purely for engagement, data extraction, or commercial advantage at a child's expense. UNESCO raised it as one of the central themes of Digital Learning Week 2026, alongside public accountability and deliberative governance.

Is a commercial AI tutoring product automatically excluded from being "public-interest"?

No, but it can't claim the label just by existing. UNESCO's framework is really aimed at governments and public education systems, so a commercial product is always a different kind of thing — and a family should judge any product, including ours, on its own specific, checkable design choices rather than on the label a company gives itself.

Duke Harewood built aitutors.me for his own KS3 daughter, and wrote this piece the way he'd want a competitor's version read: as specific, checkable claims rather than adjectives. Published 11 September 2026.