A student uses a chatbot to outline an essay and receives a confident claim about a historical event. The sentence is fluent, the citation looks plausible, and the student has little time. Should they accept it, check it, revise it, or leave the tool aside? A school can prohibit the tool or teach prompt tips, but neither response alone gives the student a way to judge the claim. The more useful educational question is what the learner must understand and remain able to do when AI participates in the task.
The OECD and European Commission’s 2026 report, Empowering Learners for the Age of AI: An AI Literacy Framework for Primary and Secondary Education, offers a common reference for that question. Released on 18 June 2026, it describes knowledge, skills, attitudes and competences for primary and secondary learners. This is a source-first guide to the released framework, its development and its limits. It does not treat a competence statement as evidence that a particular lesson or product improves learning.
What was released, and for whom
The 64-page publication is a joint OECD and European Commission framework, developed with international expert support. Its intended readers include teachers, school leaders, education authorities, learning designers and others who shape learning experiences. The European Commission’s launch announcement describes it as a common reference for integrating AI literacy, with classroom examples at primary and secondary levels. It is not a single national curriculum, a vendor certification or an assessment score for every child.
The report itself calls the framework non-binding. It says schools and systems should adapt it to their learners, subjects and context. It explicitly says it is not guidance on enforcement of the EU AI Act. This matters because the authority of a joint institutional publication can easily be mistaken for a legal command. A local school still needs to decide which tools are appropriate, what data may be used, who is responsible for teaching and how an activity fits its curriculum and applicable rules.
The framework also belongs to a particular moment. A draft appeared in May 2025; the released version followed consultation and revision. Some earlier descriptions of the draft refer to a domain called “Design AI.” The final report names the fourth domain Shape AI. A lesson plan or commentary that quotes an older draft should not silently present its labels as the final 2026 text. Version control is itself part of source literacy.
The question the framework answers
AI literacy here is not just the ability to type an effective prompt. The report frames it as knowledge, skills and attitudes that help young people understand how AI systems work, assess outputs critically, and use or shape AI ethically and creatively. A student may meet AI in a search ranking, a recommendation system, a translation feature or a conversational assistant. Some of these systems are obvious; others are built into services the student already uses. Education must therefore address recognition and judgment as well as tool operation.
The report describes desired outcomes rather than reporting the measured effect of one intervention. Its method is framework development: synthesis of existing work, expert contributions and stakeholder consultation, followed by illustrative competences and classroom scenarios. That is an appropriate method for building a shared language. It cannot by itself show that every proposed activity produces transfer to an unfamiliar tool, a lasting change in understanding or a safer decision outside school.
This distinction makes the framework more useful. A teacher can use it to specify a goal: a learner should explain why a recommendation may be biased, decide whether generated text should be accepted or rejected, or identify the information they should not enter into a tool. The teacher then needs an assessment that actually observes that ability. Counting how often a pupil opens an AI application would answer a different question.
Four domains, with the final names
The released report has 19 competences grouped into Engage with AI, Create with AI, Manage AI and Shape AI. Engage is foundational: learners recognise AI, interpret its role, assess claims and consider effects on themselves and others. Create concerns work with AI as a possible creative partner while preserving human authorship and judgment. Manage asks when to use AI, how to divide a task between person and system, and how to monitor the process. Shape moves beyond use toward examining or improving systems in light of human values and intended purposes.
The order is a proposed learning pathway, not a rule that every class must proceed through four sealed units. The report says Create and Manage can develop in parallel after a foundation in Engage. Shape may demand additional technical scaffolding and collaboration. A student can still raise a good design question before they can build a model, and a student who can code may still need help judging whether a system should be used. Technical competence and ethical judgment are related, not interchangeable.
The four domains also prevent a narrow equation of literacy with consumption. A pupil should be able to ask who selected the data, what an output is for, whose interests a tool serves and how a design could change. That ambition is different from promising that every pupil will become an engineer. The Shape domain expressly concerns age-appropriate exploration of systems and values, not production of commercial software in school.
Knowledge, skills and attitudes are not a checklist of apps
The framework separates knowledge about AI systems and their social setting from skills such as critical thinking, collaboration and problem solving, and from attitudes including curiosity, reflection and responsibility. Competence statements bring these strands together in actions a learner might show. The report adds learner expectations and classroom scenarios, with progressions described as basic, intermediate and advanced. Those levels are not fixed to an age or grade; learners arrive with different experience and opportunities.
Consider the essay claim. Knowledge includes understanding that a generative system can produce plausible text without checking a historical archive. Skill includes tracing a claim to a reliable source and revising one’s own argument. An attitude of responsibility includes telling a reader where AI contributed and not presenting an unchecked citation as verified. A student who knows the vocabulary but cannot apply it has not yet shown the whole competence. A student who happens to catch one falsehood may still need a more reliable process.
The framework’s scenarios should be treated as examples for design, not as scripts that override a teacher’s expertise. A teacher might use a pre-generated answer so pupils can critique it without creating accounts or disclosing data. Another might compare a system’s suggestion with a local source in a language the system handles poorly. The relevant question is whether the task exposes the intended reasoning and whether pupils can explain their decisions. Brightly worded scenarios are not a substitute for that observation.
What changed after consultation
The report states that more than 2,000 individuals gave feedback on the draft through an online survey, focus groups and written reviews, with participation from over 100 countries. Teachers made up a substantial share of the survey respondents. Ministries of Education from several EU countries, including Denmark, Finland and Sweden, provided written reviews. The authors say the feedback helped them make the material more adaptable and practical.
That process broadens the perspectives informing a framework; it is not a randomized evaluation of the final text. The distribution of respondents was not designed to represent every teacher, student or country equally. The report names youth among participant groups, but their share of the feedback survey was small. A school should therefore invite its own learners, families and staff into implementation decisions rather than assuming an international consultation has already captured its particular barriers.
The revision history matters in another way. AI systems and their use change quickly, while education systems move through slower curriculum, procurement and professional-development cycles. A framework can offer durable questions about evidence, data, accountability and agency even when a named product changes. Yet it still needs periodic review. This policy brief should be revisited when the OECD or Commission changes the report, publishes new exemplars or issues materially different guidance.
Ethics runs through the domains
The report does not place ethics in one isolated lesson to be completed before returning to ordinary tool use. It weaves privacy, fairness, accountability, environmental stewardship and attention to who benefits throughout the framework. A student judging an AI recommendation should consider whether it excludes a perspective or repeats a stereotype. A student generating an image should consider provenance and the rights of people whose work shaped it. A student managing a workflow should know when sensitive information should stay out of a system entirely.
These questions cannot be answered by slogans such as “be responsible.” They require context and evidence. For example, a system’s energy use depends on its design and infrastructure; a school should not present a single universal number for every query. A bias claim should identify the task, output and affected group. Privacy depends on the actual provider, contract and data flow. The framework supplies a reason to investigate these matters and a language for instruction, not a finding that any particular product has passed a safety review.
Human agency is central, but it should not be reduced to making one pupil carry every risk. A learner can decide whether to accept a suggestion only if they can see enough to evaluate it and have permission to reject it. Teachers need time, training and appropriate materials. School leaders must address procurement, accessibility, data handling and classroom rules. The report explicitly distributes responsibility across the education ecosystem. This is a practical design requirement, not merely a moral appeal to individual caution.
How this differs from related frameworks
UNESCO’s 2024 AI Competency Framework for Students outlines twelve competencies across four different dimensions, including a human-centred mindset, AI ethics, techniques and applications, and system design. It is a distinct publication with its own progression model. The 2026 OECD/Commission AILit report draws on UNESCO and other frameworks while organizing its own school-facing domains and scenarios. Treating their categories as if they were the same numbered standard would obscure what each source actually says.
The European Commission’s DigComp 3.0 is broader digital competence guidance. Its Joint Research Centre description integrates AI competence across a wider set of digital activities, including information literacy, communication, content creation, safety and problem solving. AILit narrows the lens to AI literacy in primary and secondary education. The Commission also has 2026 ethical guidelines for educators on AI and data use in teaching and learning. Those address teacher and institutional decisions, while AILit describes learner outcomes. A school may use these documents together without pretending that one replaces the others.
The OECD’s planned PISA 2029 Media and Artificial Intelligence Literacy assessment is separate again. The OECD describes that project as a broader assessment of media and AI literacy and, as of October 2026, labels its assessment framework a preliminary draft. AILit contributes to that work, but the framework does not itself report a PISA score or validate a school against an international benchmark. The earlier UNESCO guidance for generative AI in education and research is also a different kind of document: a 2023 policy and governance guide, not the final 2026 school-literacy framework.
The learning-evidence question remains open
A framework can describe what students should learn without proving which teaching strategy will produce it. The OECD Digital Education Outlook 2026 reviews emerging evidence that general-purpose generative AI can improve the quality of an assisted task without producing the same gain in independent learning. Purposeful, scaffolded use can have better prospects, but effects depend on subject, learner, tool and assessment. The AILit report itself acknowledges that evidence of generative AI’s efficacy for learning is mixed.
This is why a school should distinguish performance with assistance from understanding after assistance is removed. A polished essay may show that a pupil selected and edited a useful output. It may not show that they can locate a source, construct an argument or revise a claim without the tool. An evaluation could compare assisted work, later independent explanation and the student’s account of when they accepted or rejected suggestions. It should also test whether any gains persist beyond a single lesson.
The reverse assumption is equally weak. A ban on every AI interaction does not demonstrate stronger judgment about systems that pupils meet outside school. A lesson on critical evaluation may be possible without letting children submit personal data to an external service. The evidence question is not “AI or no AI” in the abstract. It is what learning goal is at stake, what exposure is necessary, what alternative design is available, and what outcome would count as success without transferring hidden costs to learners.
Assessment that can see a decision
The framework’s competence language invites performance tasks. For Engage, a student could explain why a source-like AI answer is insufficient and show how they would verify its claim. For Create, they could annotate which parts of a work are their own and where a tool contributed, then justify each adopted suggestion. For Manage, they could describe a task they chose not to delegate and explain why. For Shape, they could compare a system’s behavior across test cases and propose a change based on the observed failure.
Such tasks are proposals for local assessment, not validated tests supplied by this article. A teacher should decide what counts as sufficient evidence at the learner’s stage. Scoring only the finished product risks missing the very judgment the framework values. Scoring only a student’s confidence risks mistaking fluency for understanding. Recording reasons, revisions and counterexamples can make the process more visible, although no single rubric captures every kind of AI literacy.
Fairness also matters in assessment. Students differ in device access, language, disability support and opportunities to practice outside school. A test that rewards familiarity with one commercial interface may measure access as much as judgment. A school can use common examples, alternative formats and explicit instruction before assessment. It should evaluate whether a learner can transfer a principle to a new case, rather than only remember the interface practiced last week.
Why this matters in Northern Europe
The report’s consultation included written input from ministries in Denmark, Finland and Sweden, alongside other European countries. That is a relevant Nordic link, but it does not mean the framework has already been adopted as a national curriculum in each country. Curriculum authority, language rights, privacy rules, procurement and teacher development differ. The final report is explicitly adaptable and non-binding, so national and local choices remain necessary.
In a multilingual classroom, a learner may need to test whether an AI system handles Swedish, Finnish, Danish, Sámi or a home language with the same reliability as a dominant training language. The framework’s concern with evaluation and inclusion supports that question, but the answer requires local tests of specific tools. It would be careless to infer equal quality from the existence of a general framework. Schools also need ways for students to learn the principles without buying access to a particular commercial system.
The Institute’s synthesis is that AI literacy is part of a wider practice architecture: time to check a claim, an adult who can discuss uncertainty, clear rules for data and attribution, and assessment that rewards reasoning. A framework gives shared terms for that architecture. It does not supply staffing, resolve unequal access or establish that every pupil has the same freedom to challenge a machine-generated answer. Those conditions must be built and evaluated by the institutions using it.
What the evidence does not show
The 2026 AILit Framework does not show that its four domains have been validated as a universal psychometric scale, that a school implementing them will raise attainment by a known amount, or that the classroom scenarios are interventions with proven long-term effects. Its consultation informed the framework but did not randomly assign schools to competing curricula. It does not determine whether a particular child’s use of AI helps or harms learning. Those questions require outcome studies with clear comparators, time horizons and attention to equity.
It does not certify a provider, create a binding obligation to use AI in every subject, or explain how to enforce the EU AI Act. Nor is it identical to UNESCO’s student competencies, DigComp 3.0 or the preliminary PISA 2029 assessment design. Its strongest claim is more modest and more useful: education systems now have a current, shared vocabulary for the knowledge, skills, attitudes and decisions they want young people to develop. Whether that vocabulary changes practice will depend on local design and subsequent evidence.
Back to the uncertain citation
The student returns to the fluent sentence. They may reject it, verify it in a historical source, or use it as a prompt to ask a better question. A good AI-literacy lesson would let them explain why they chose one of those paths, and would make it possible to choose not to use the tool. That is a small moment, but it contains the framework’s larger ambition: informed participation rather than passive acceptance.
The released OECD and European Commission report offers a practical map for that ambition, not proof that the journey is complete. Its value lies in the questions schools can now ask with greater precision: what should a learner recognise, decide, create, manage or reshape; what support makes that possible; and what evidence would show that the ability remains when the interface changes?
Research references
- OECD and European Commission. (2026). Empowering Learners for the Age of AI: An AI Literacy Framework for Primary and Secondary Education. OECD Publishing. View source
- European Commission. (2026, 18 June). New AI Literacy Framework helps schools prepare learners for the age of artificial intelligence. View source
- OECD. (2026). OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education. OECD Publishing. View source
- UNESCO. (2024). AI competency framework for students. View source
- European Commission Joint Research Centre. (2025). DigComp 3.0: European Digital Competence Framework. View source
- OECD. (2026). PISA 2029 Media and Artificial Intelligence Literacy project and preliminary assessment framework. View source
- European Commission. (2025). Empowering learners for the age of AI: Launch of the draft AI literacy framework and stakeholder consultations. View source