More often than not, conversations about artificial intelligence in language education tend to revolve around practical questions: Which tool works best for which tasks? How can I write more effective prompts? How can I use AI to streamline my work processes? These are reasonable questions, and they deserve good answers. But, I think that they stop short of the questions that shape practice in more fundamental ways.

What does it mean to use AI well in a language classroom?

The word ‘well’ above does a lot of work. It means not just effectively, but ethically. That is, in ways that are legally sound, that respect learners, that do not quietly reproduce bias or erode teacher agency, and that add genuine educational value rather than simply adding superficial novelty.

This is the question that we have been working to answer at the AI Lang project, a four-year initiative of the European Centre for Modern Languages of the Council of Europe. One of the main outputs of this project is a framework of guidelines for the ethical use of AI in language education, and I have traced its development in this blog over the last two years. As we are now close to finalising the guidelines, I’d like to say a few words about what this framework contains, how it is structured, and where you can explore it for yourself.

What we have arrived at is not a set of easy instructions, but a structured way of thinking about practice. In what follows, I unpack how we have approached this question, and what has emerged from that process.

The AI Lang framework

As you are probably aware, there is no shortage of AI guidance documents in circulation at the moment. What distinguishes the AI Lang framework, in the most obvious sense, is its specific focus on language education. More deeply, and I would argue more importantly, the AI Lang framework also differs because we approach ethics not as a compliance checklist but as a set of questions that teachers need to be able to ask for themselves.

Structure of the framework

The basic building blocks of the framework are four foundational principles. The first one is that AI use in language education should be safe, meaning that it should be legally compliant, transparent, and risk-aware. Secondly, AI use in language education should be responsible, in the sense that it positions itself as a force for positive change. Thirdly, it should be purposeful, by which we mean that it should add genuine pedagogical value to teaching and learning. And finally, it should be reflective: supporting teacher agency and professional growth.

Each principle is operationalised through two specific guidelines, and each guideline is further specified by a set of competence descriptors. These are concrete, observable indicators of what teachers and learners do when they enact a principle in practice. For each descriptor, there are several reflection questions that encourage teachers to translate this guidance into practice that makes sense in their own context. Across the full framework, there are eight guidelines and thirty-five competence descriptors in total.

Safe use of AI in language education

The safety principle encompasses two guidelines: (a) compliance with legal requirements and institutional standards, and (b) the safeguarding of data security and digital safety. In practice, this means things like understanding what data AI tools collect, verifying that tools comply with relevant data protection legislation, and protecting vulnerable learners (e.g., minors, migrants, refugees) from potential data misuse. It also means staying current with regulation: the EU AI Act, for instance, prohibits the use of emotion recognition and real-time facial recognition in schools, which has direct implications for how teachers evaluate and adopt AI tools for educational use.

Responsible use of AI in language education

The responsibility principle takes us from legal compliance into the domain of values. Its two guidelines address social justice and linguistic diversity on one hand, and environmental sustainability on the other. Social justice and linguistic equality are perhaps are perhaps not so surprising to those who have been following my work. But the inclusion of environmental sustainability as a named guideline was a deliberate decision and, in our view, important topic to address: the carbon, water, and energy footprint of large-scale AI use is debatable but real, and a framework that ignores it would be incomplete. A practical implication is that teachers should develop habits of calibrated, purposeful AI use: they should reach for AI when it genuinely adds value, rather than as a default.

Purposeful use of AI in language education

Pedagogy comes to the foreground with the purposeful principle. The framework’s fifth and sixth guidelines focus respectively on enhancing sound language teaching practice and on creating novel learning affordances, i.e., opportunities that go beyond what was previously possible. Sound pedagogical practice encompasses all the work that one might expect: meaning-focused work, context-sensitive pedagogy, and more. For novelty, a useful reference point would be the SAMR model, which distinguishes between AI use that merely substitutes for existing tasks and AI use that genuinely transforms what is possible. The framework pushes toward the transformative end of that spectrum, while insisting that pedagogical intent —not technological novelty— should drive decisions.

Reflective use of AI in language education

The reflective principle, which covers teacher empowerment and professional agency, and it is something which I will be writing about in more detail in coming weeks. The framework’s position is that teachers should move from being end-users of AI tools to being co-designers of AI policies and practices. In other words, they need to be participants in shaping how AI enters their professional contexts, not simply recipients of decisions made elsewhere.

Exploring the framework interactively

The AI Lang team has developed an interactive version of the framework, currently in beta, which presents the four principles, eight guidelines, and thirty-five competence descriptors in a navigable format.

This is, to be transparent, a beta release. For the time being, we are hosting it in my university drive, until we are ready to move it to a more permanent home. We are also aware that has some formatting rough edges. However, we are sharing it now because we think the content may be useful to some and because we genuinely want feedback, not only on the framework itself, but on the usability of the interactive format. If something is unclear, hard to navigate, or if we could present something more effectively, we would very much welcome your thoughts in the comments below or via the contact details on the site.

What I think that the interactive format does well, even in its current state, is make visible the internal logic of the framework, i.e., the way principles connect to guidelines, and guidelines connect to specific competences. It also includes a few practical asides, like the summary of what the EU AI Act means for schools, notes on AI bias in current language models, a prompt design framework (IDEA) for effective AI interaction, and a sample reflective prompt for post-lesson analysis. We hope that these make the framework more usable rather than merely more readable.

Validation and next steps

A framework of this kind is only as good as its grounding in the experience of the people it aims to serve. At the moment we are working in a validation process, consulting with experts in language education to assess how clearly we have articulated the competence descriptors and how relevant they are in practice. That work is ongoing, and we hope to complete it in the upcoming network meeting in Graz (9-10 April 2026).

In the meantime, however, the framework is available to explore, to use, and to respond to. If you work in language education —as a teacher, a teacher educator, or a policy planner— and you have thoughts about what the framework gets right, what it misses, or how it could be more useful in your context, those responses are valuable. The comments section below is one place to share them.

---