Ei-iE

What Do AI Frameworks for Education Really Look Like?

The capabilities of Artificial Intelligence (AI), and in particular Large Language Models like ChatGPT, Claude, or DeepSeek, are developing at extraordinary speed, accompanying by calls from some tech firm leaders to slow the pace and ask serious questions about regulation. Paralleling these developments have been a dizzying number of indexes, frameworks and indicators all aimed at AI, whether oriented to AI readiness, AI literacies, or AI competences.

Good, we might reassure ourselves. When it comes to education systems, we have a duty of care, and thus responsibility, to learners, teachers, and school administrators. Such frameworks should serve this purpose. But do they? What do they look like when scrutinized closely, and who are they really aimed at? Promoting more AI use to feed the machine with more and more data traces? Or driving deeper than that and putting into place the guardrails to ensure use with safety and ethics?

For this reason, in our recently published research in the European Education Research Journal, we examine the AI literacy frameworks being promoted by UNESCO and the OECD/European Commission. These frameworks are particularly relevant because international organizations have considerable power to shape national education policy. A case in point is the OECD/EC framework, which is intended to inform the assessment of Media and AI Literacy in PISA 2029.

In our study we analysed three major policy documents aimed at national education systems around the globe: UNESCO’s AI competency frameworks – one for students and the other for teachers, both published in 2024, and the OECD/EC AI literacy framework published in 2025. Using qualitative policy analysis alongside text-mining and network analysis, we examined which actors appear in these documents, the relationships between them and, crucially, the kind of agency attributed to students and teachers. What we found should give us pause for further thought.

Regulation without a regulator

The first striking finding concerns regulation. All three frameworks recognise the legal and regulatory challenges surrounding AI. In some cases, they emphatically underscore the point. And yet, our analysis found that governmental institutions are largely absent. Most strikingly, the OECD/EC document does not mention a governmental institution at all. In short, the frameworks call for regulation, as a sort of mantra, without clearly identifying who has the authority, or responsibility and power, to regulate. As we put it in our paper, this is “ an action without an actor”. That absence matters.

Schools have responsibilities towards children and young people. Yet they are being encouraged to integrate rapidly developing frontier technologies pushed by tech oligarchs whose data practices, environmental costs, biases, commercial interests and wider social consequences remain deeply contested. In this context, repeated calls for regulation seem a way to hide the problem in plain sight. The question remains: who will decide where the limits lie?

Human-centred or AI-centred?

The second finding is a paradox.

These frameworks repeatedly use the language of “human-centred” AI. Yet when we analysed which actors actually dominate the documents, we found something very different. AI is not just extraordinarily prominent, it is the main actor in the frame.

In the OECD/EC framework, AI is almost the only actor discussed independently; as a result, it is a subject endowed with its own agency. By contrast, students and teachers appear overwhelmingly only in relation to AI. UNESCO’s student framework is somewhat more balanced, while its teacher framework gives greater recognition to the teacher-student relationship.

This raises a simple question: how human-centred is a framework in which the human actors are primarily defined through their relationship with AI? The answer becomes more troubling when we look at students.

From student to developer

To understand how future students are imagined to act and live in the so-called age of AI, we analysed the verbs associated with students in these frameworks. They include develop, build, engage, try, apply, leverage, refine and evaluate. Others include construct, explore, assess and create. This is, to say the least, an unusual educational vocabulary.

Our view is that the agency attributed to students resembles that of a developer more than that of a traditional student. Rather than primarily encountering bodies of knowledge, students are imagined as acquiring competencies through continuous interaction with AI: experimenting, testing, assessing, creating and refining. This changes the educational relationship.

That is, the development of students' capabilities and their continuous interaction with AI systems become part of the same process. Students increasingly appear as cybernetic co-producers operating within feedback loops between humans and machines. The only knowledge that matters is that which is locally employed towards the production of a practical outcome. How the value of this product will be assessed remains dangerously unexplained.

This also brings us to the question of data. Every interaction with an AI system generates traces that are then used to develop and subsequently monetize AI systems. At scale, education represents an enormous field of human-machine interaction. Capture this sector, and the technology firm that wins has a huge global market for AI products. More than this, once introduced as an infrastructure that underpins learning with AI, schools and universities then become the ongoing testbeds for refinements as well as new innovations. Given this kind of scenario, we argue that we need to think about education not simply as a site for the consumption of AI, but as a site of value extraction.

And what happens to teachers?

If students become increasingly active in relation to AI, the position of teachers looks very different. In the OECD/EC and UNESCO student documents, teachers barely appear as meaningful independent actors. Even in UNESCO’s teacher framework, their agency is peculiar. Teachers need things. They need to be trained, equipped, encouraged, helped, supported, prepared, guided and empowered. They are, paradoxically, mostly on the receiving side of a ‘treatment’ to enable the new AI-student relation to form.

When teachers do act, they co-design, co-create, develop, safeguard, demonstrate and integrate. Strikingly, our textual analysis found that they are not principally entrusted with teaching. The result is a curious reversal: students are imagined as developers engaging directly with AI, while teachers become auxiliary figures who require training to facilitate that relationship.

In this framing, teachers are to help students along, rather than mediating or (god forbid) limit their relationship with AI. On the contrary, we argue in these frameworks a teacher needs to be imagined and represented as someone who must raise important questions around power, interests, bias, sustainability, and ethics. A key role for the teacher is to help students understand the broader political economy of these tools and the firms. Asking how the large technology firms are profiting significantly from the students’ classroom labouring makes visible their business models.

Adding education to technology

We also see a much bigger issue here. So far, technological innovation has entered education through some version of this sequence: a technology is developed, tested within limited settings (e.g. pilots or beta versions), debated, regulated, modified and eventually commercialized. In short, tech companies used to view education as a market in which to sell specifically designed educational technologies.

Generative AI reverses that relationship. Instead of asking how technology might be carefully incorporated into education, education is increasingly being incorporated into the development of AI. That distinction is fundamental.

If students continuously interact with proprietary AI systems, teachers facilitate those interactions, and schools provide the institutional environment in which they occur, then education systems risk becoming live testing grounds for AI. And if those interactions generate economically valuable data traces, they potentially (and likely) become sites of extraction too.

AI literacy cannot simply mean preparing teachers and students to adapt to a technological future that has already been decided for them, where “preparing” often means to make sure there will be no significant resistance against the acceptance of such technologies. It must also mean understanding the wider political economy of AI by asking: who owns the infrastructure, who designs the systems, who captures the data, who benefits, who regulates, and who has the power to refuse when the red lines are overstepped?

Before we ask how education should adapt to AI, shouldn’t we be asking a more fundamental question.

Surely it is us, as educators and citizens (and not the commercial firms), who get to decide what is educationally relevant, what does safety look like, what are the responsibilities of the teacher in the classroom with students and AI, and how might governments and independent auditors regulate and report on these services?