UNESCO vs OECD vs EU AI Act: Where International AI Frameworks Actually Disagree
Three frameworks, three starting points, three blind spots
Alex Gray
Director, DEEP Education
When I first started building the AI Literacy Audit Tool, I assumed the hard part would be technical. Getting the assessment logic right, designing the scoring, building the interface, would take months. I was wrong. The hard part was working out what schools should actually be measured against. Because when you sit down and read 33 international frameworks on AI in education, you quickly discover something uncomfortable: they don't all agree.
Not on the big stuff. Not on the details. Not even on what "AI literacy" means.
Most school leaders I speak to have heard of one or two of these frameworks. Maybe UNESCO's guidance. Perhaps a nod to the EU AI Act. Occasionally someone mentions the OECD. But almost nobody has read them side by side, and that's where things get interesting, because the gaps between them tell you as much as the frameworks themselves.
The Three Heavyweights
Let me focus on the three that come up most often in conversations with international school leaders: UNESCO's AI Competency Framework for Teachers, the OECD's AI Principles as applied to education, and the EU AI Act.
They come from very different places. UNESCO is a normative body, it sets aspirational standards. The OECD is an economic cooperation organisation, it thinks in terms of productivity, innovation, and human capital. The EU AI Act is legislation; it carries legal force and compliance obligations. Those origins shape everything about what they prioritise.
UNESCO starts with teachers. Its competency framework is arguably the most classroom-facing of the three. It defines what teachers need to know, understand, and be able to do with AI. It talks about pedagogy, ethics, and human oversight. It is explicitly values-driven: equity, inclusion, and cultural sensitivity run through every section.
The OECD starts with systems. Its AI principles, originally published in 2019 and updated since, focus on trustworthy AI at scale. When the OECD talks about education, it talks about workforce readiness, skills pipelines, and national competitiveness. It is not unkind to classrooms, but it views education as an input to economic output.
The EU AI Act starts with risk. It classifies AI systems by their potential for harm, and educational AI sits firmly in the "high-risk" category. That means if your school uses AI for assessment, student profiling, or admissions decisions, you are operating under the most stringent regulatory tier. The Act does not tell you how to teach with AI. It tells you what you must prove about the AI systems you deploy.
Where They Converge
There is genuine overlap. All three agree that transparency matters, students and teachers should understand when AI is being used and broadly how it works. All three agree that human oversight is non-negotiable, AI should support human decision-making, not replace it. And all three agree that data protection is foundational.
If your school gets these three things right, transparency, human oversight, and data protection, you have a baseline that holds up across frameworks.
Where They Diverge
Here is where it gets complicated.
On teacher competency: UNESCO says teacher AI literacy must come before student AI literacy. It is explicit about this: the framework positions teacher readiness as a prerequisite, not a parallel track. The OECD does not take this position. It treats teacher upskilling as one of many systemic interventions, alongside curriculum reform, infrastructure investment, and assessment redesign. The EU AI Act does not mention teachers at all. It cares about the technical compliance of AI systems, not the pedagogical readiness of the people using them.
This matters enormously for school leaders. If you follow UNESCO, you invest heavily in CPD before rolling out AI tools to students. If you follow the OECD, you might pursue a broader systems approach. If you follow the EU AI Act, you focus on procurement and vendor compliance. Three frameworks, three very different starting points.
On student agency: UNESCO foregrounds student voice and participation. Students should not just be subjects of AI; they should understand it, question it, and shape how it is used in their learning. The OECD acknowledges student agency but frames it in terms of future skills and employability. The EU AI Act frames students as data subjects with rights, but not as active participants in AI governance.
For an international school trying to build a student AI literacy programme, these are meaningfully different philosophical positions.
On assessment: This is the sharpest divergence. UNESCO barely touches assessment; its framework is about teaching competencies, not assessment integrity. The OECD recognises that AI disrupts traditional assessment but treats it as a design challenge for education systems to solve over time. The EU AI Act, by contrast, puts AI-driven assessment squarely in the high-risk category and demands conformity assessments, documentation, and human oversight mechanisms with teeth.
If your school uses any form of AI in assessment, even AI-assisted marking or plagiarism detection, the EU AI Act has direct implications that neither UNESCO nor the OECD address with the same specificity.
On data and privacy: All three care about data protection, but the depth varies dramatically. The EU AI Act inherits the full weight of GDPR and adds AI-specific requirements on top. UNESCO offers principles but not compliance mechanisms. The OECD sits somewhere in between, recommending robust data governance without the legislative force to require it.
For schools operating across jurisdictions, which is essentially every international school, this creates a compliance puzzle. Which standard do you meet? The answer, in practice, is that you meet the highest one, because if you satisfy the EU AI Act you will likely satisfy the others. But most schools are not even aware that the EU AI Act classifies educational AI as high-risk.
What Most Schools Are Missing
In the hundreds of audits we have processed through the AI Literacy Audit Tool, I see the same pattern repeatedly. Schools anchor to one framework, usually UNESCO, because it is the most accessible, and assume they have covered their bases. They have not.
The most common gaps are in three areas. First, assessment governance: schools have no documented process for evaluating AI tools used in or near assessment. Second, cross-jurisdictional compliance: schools operating in the UAE, for instance, are subject to different data protection regimes than schools in the UK or EU, and their AI policies rarely reflect this. Third, teacher readiness measurement: schools claim to be investing in CPD but have no way of measuring whether teacher AI competency is actually improving.
These gaps exist because no single framework covers everything. You need the pedagogical depth of UNESCO, the systemic thinking of the OECD, and the compliance rigour of the EU AI Act. And then you need to reconcile them.
What to Do About It
If you are a school leader reading this, here is my practical advice.
Start by mapping your current AI use against the EU AI Act's risk categories. If anything you use falls into high-risk, and if you use AI in assessment or student profiling, it almost certainly does, you need compliance documentation that most schools currently do not have.
Then layer UNESCO's teacher competency framework on top. Use it to audit your CPD provision. Are your teachers actually developing AI competency, or are they attending one-off sessions that tick a box? There is a difference, and UNESCO's framework helps you see it.
Finally, use the OECD's systems lens to check your governance structures. Do you have an AI steering group? Is AI part of your strategic plan? Can you answer the board's questions about AI with evidence rather than anecdote?
The AI Literacy Audit Tool was built to do exactly this: to cross-reference your school's readiness against all 33 frameworks simultaneously, so you do not have to read them all yourself. But whether you use the tool or not, the principle stands: one framework is not enough. The international landscape is fragmented, and schools that anchor to a single source will have blind spots.
The question is not which framework is right. They all are, within their own logic. The question is whether your school has the structures, skills, and evidence to satisfy the full picture. In my experience, most do not. Yet.
Alex Gray
Director, DEEP Education
Education technology specialist with 20 years in the education sector. BSME AI Network Lead and ISC Edruptor 2024 & 2025. Alex founded DEEP Education, part of the DEEP Education Network by DEEP Professional, to help schools navigate AI integration with confidence.
Ready to assess your school’s AI readiness?
Upload your policy documents and receive evidence-based scores across all 9 dimensions, with actionable improvement plans.
Start your free audit