AI Literacy Frameworks Compared: Which One Should Your Institution Adopt in 2026?
You’ve read three summaries of three different AI literacy frameworks this month and you still can’t tell which one to actually pick. That isn’t your fault. Most articles on this topic list the frameworks and describe what each one emphasises, but they don’t put them on consistent axes that a decision maker can act on.
This piece does. I’ve worked with two of these frameworks directly while building Lumination’s AI literacy self-assessment for educators, and looked closely at the other three while choosing what to build on.
What follows is a side-by-side read on the five most-referenced AI literacy frameworks in education today: where they came from, what shape they take, and the situations where each one earns its place.
If you want the verdict before the detail, the comparison table is in the next section. EU AI Act Article 4 has been in force since 2 February 2025, which makes the choice more load-bearing than it was a year ago. If your wider question is whether the AI tools your school already uses are themselves classified as high-risk under the Act, that’s a separate question covered by our free EU AI Act risk assessment tool.
What Actually Differs Across AI Literacy Frameworks (The Comparison Table)
Most coverage of AI literacy frameworks treats them as if they’re variations of the same thing. They aren’t. The field has converged on broad goals (critical thinking, ethics, responsible use, human agency), but the frameworks diverge sharply on properties that matter for implementation: whom they’re for, whether they translate into assessable items, and whether they’re EU AI Act-aligned by design.
Here is the side-by-side, on seven axes that affect real decisions.
| Framework | Governance | Year | Audience | Structure | Assessment-ready? | EU AI Act Art. 4 fit | Free? |
|---|---|---|---|---|---|---|---|
| OECD-EC AILit (Empowering Learners) | OECD + European Commission | Draft 2025, final 2026 | Teachers, leaders, policymakers, learning designers (K-12) | Domains TBD in final draft | Partial (exemplars pending) | Strong (co-developed with EC) | Yes |
| UNESCO AI Competency Frameworks | UNESCO | 2024 | Teachers and students (paired frameworks) | 5 dimensions × 15 competencies × 3 levels (teachers); 4 aspects × 3 levels (students) | High (explicit mastery levels) | Moderate (not EU-specific) | Yes, multilingual |
| AI4K12 | AAAI + CSTA | 2019 (active) | K-12 curriculum developers and teachers | Five Big Ideas across grade bands | Low (curriculum guidance, not assessment) | Low (no EU connection by design) | Yes |
| ETS AI Literacy Framework | ETS | 2025 | K-12 students | Conceptual framework, learning progression, task design principles | High (built for assessment) | Moderate (US-leaning) | Yes (CC license) |
| Long & Magerko | Academic (CHI 2020) | 2020 | Researchers, framework designers | 5 overarching + 17 sub-competencies + 16 design considerations | Medium (analytical, not assessment) | Predates the Act | Yes (open access) |
Three of these frameworks (OECD-EC, UNESCO, ETS) can actually be translated into staff training programmes and assessment items today. AI4K12 sits upstream of that work, shaping what computer science teachers teach. Long & Magerko sits behind almost all of them as the academic backbone, which is why you keep seeing it referenced.
The rest of this article goes through each one in turn, starting with the framework most people mean when they say “the EU AI Literacy Framework”.
1. OECD-EC AILit Framework (The “EU AI Literacy Framework” You’ve Been Hearing About)
If you’ve seen this framework called the EU AI Literacy Framework, the OECD AI Literacy Framework, the EC AI Literacy Framework, the Empowering Learners Framework, and the AILit Framework, you’ve seen five names for the same document.
The official name is Empowering Learners for the Age of AI: An AILit Framework for Primary and Secondary Education. It is a joint initiative of the OECD and the European Commission, with support from Code.org.
The draft was published in May 2025. The final version is scheduled for 2026 alongside concrete “AI literacy exemplars” that translate the competencies into classroom items.
The framework addresses four named groups: teachers, education leaders, policymakers, and learning designers. The focus is primary and secondary education.
The draft is open for stakeholder consultation, and the precise domain count is still settling as that consultation runs. The framework explicitly flags “lack of a shared understanding of AI literacy” as a starting problem it intends to solve.
Lumination’s own free AI literacy self-assessment for educators is grounded in this framework alongside UNESCO’s teacher framework. The two pair well because the OECD-EC document sets the longer-term map while UNESCO supplies the immediately operational structure.
One thing the draft does well that earlier frameworks didn’t: it speaks to four audiences (teachers, leaders, policymakers, learning designers) rather than collapsing them into “educators”. That separation matters because the obligations Article 4 creates for an education leader (procurement, vendor due diligence, staff training plans) are not the obligations it creates for a classroom teacher.
2. UNESCO AI Competency Frameworks (Teachers and Students, 2024)
UNESCO published the two AI literacy frameworks most ready to use today. One is for teachers, one is for students. Both are free, multilingual, and structured around explicit mastery progression levels, which is what makes them implementation-ready in a way most of the others aren’t.
The UNESCO AI Competency Framework for Teachers was released on 8 August 2024. It is structured as 5 dimensions × 15 competencies × 3 progression levels (Acquire, Deepen, Create). The five dimensions are: human-centred mindset, ethics of AI, AI foundations and applications, AI pedagogy, and AI for professional learning.
The parallel UNESCO AI Competency Framework for Students uses 4 aspects (human-centered mindset, ethics of AI, AI techniques and applications, AI system design) and the same three mastery levels (Understand, Apply, Create), giving 12 competency blocks in total.
Both documents are available in Arabic, English, French, Portuguese, Spanish, and Vietnamese, which matters if your institution operates across language regions or trains international staff.
The progression levels translate directly into CPD rubrics. The mastery levels look like what CPD coordinators already work with, and the 15 teacher competencies are concrete enough that you can write workshop briefs against them without further abstraction.
3. AI4K12: The Five Big Ideas of AI (AAAI + CSTA)
AI4K12 is older than the AI literacy conversation it now appears in. The Five Big Ideas of AI poster was released on 16 April 2019, by a joint initiative of the Association for the Advancement of Artificial Intelligence and the Computer Science Teachers Association. The work continues, but the core structure has been stable for several years.
The Five Big Ideas of AI are:
- Perception. Computers perceive the world through sensors.
- Representation & Reasoning. Agents maintain representations of the world and use them to reason.
- Learning. Computers can learn from data.
- Natural Interaction. Intelligent agents need many kinds of knowledge to interact with humans.
- Societal Impact. AI applications can impact society in both positive and negative ways.
The audience is K-12 educators, curriculum developers, and standards writers, not directly students. The structure is grade-band guidance rather than competencies or mastery levels.
It exists to inform what computer science classrooms cover, which is a different job from the one OECD-EC and UNESCO are trying to do.
Compared to OECD-EC or UNESCO, AI4K12 sits upstream. You don’t pick AI4K12 instead of UNESCO’s teacher framework.
You use AI4K12 to inform what your computer science teachers actually teach, and then use UNESCO’s teacher framework to develop those teachers as practitioners. The two coexist, and that pairing is closer to how the field assumes the frameworks will be used than any single-framework adoption is.
4. The ETS AI Literacy Framework for K-12 Students (2025)
Most AI literacy frameworks describe what students should know. The ETS framework, by design, also specifies how you’d measure whether they actually know it.
Published in 2025 as part of the ETS Research Report Series, Preparing K-12 Students With AI Literacy: Proposed Framework, Progression, and Task Design Principles by Srijita Chakraburty, Teresa Ober, and Lei Liu has three explicit components:
- A conceptual framework
- A hypothesised learning progression
- Assessment design principles
The audience is K-12 students. The competency areas covered:
- Foundational knowledge
- Societal implications
- Practical applications
- Ethical decision-making
- AI-powered collaboration
- Critical evaluation of AI outputs
The framework is free under a Creative Commons license. Note that ETS also offers a separate product called Praxis Futurenav Adapt AI, which assesses teachers. It’s a commercial offering, distinct from the open research framework. The two are sometimes confused in coverage, which is worth flagging if your procurement team starts asking questions.
ETS’s contribution is US-leaning by origin and compatible with EU AI Act requirements, but it was not co-designed for them. If your institution sits outside the EU, that origin question matters less.
5. Long & Magerko (CHI 2020): The Academic Framework Behind Everything Else
Most of the frameworks above stand on the shoulders of one academic paper. Duri Long and Brian Magerko’s 2020 paper What is AI Literacy? Competencies and Design Considerations is the academic backbone of the entire applied field. If you’ve read three other AI literacy frameworks and noticed they all seem to ask similar core questions, this is why.
The paper was published in the proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, and is openly accessible.
The structure has three layers: five overarching competencies, seventeen detailed sub-competencies, and sixteen design considerations for educators building AI literacy programmes. The five overarching competencies are framed as questions the AI-literate person can answer:
- What is AI?
- What can AI do?
- How does AI work?
- How should AI be used?
- How is AI perceived by humans?
The paper is sometimes informally referenced under the “ABCs of AI” framing in educator materials, but the canonical citation is Long & Magerko 2020.
The audience is researchers, framework designers, and curriculum architects, not classroom teachers. You probably won’t roll out Long & Magerko in your school. But if you adopt UNESCO’s framework or OECD-EC and want to understand why each is structured the way it is, this paper is the upstream source.
The Bottom Line
The right framework depends less on which one is “best” than on what job you’re hiring it for. Three situations cover most institutions.
Article 4 compliance is on your radar
Pair OECD-EC AILit (future-state map, finalising 2026) with UNESCO Teachers (trainable competencies you can run this term). Don’t wait for the OECD-EC final release before starting.
You need measurable outcomes this term
Pick UNESCO if you need teacher progression. Pick ETS if you need student outcome measurement. UNESCO maps to CPD, ETS maps to assessment items. They’re also compatible if you want both.
Writing or revising your CS curriculum
AI4K12 sits upstream of all of this. Use it to shape what gets taught in CS class. Then layer UNESCO Teachers on top for staff CPD.
FAQ
Is the EU AI Literacy Framework the same as the OECD AI Literacy Framework?
Yes. The framework hosted at ailiteracyframework.org is jointly published by the European Commission and the OECD, with support from Code.org. Both names refer to the same document. The official name is Empowering Learners for the Age of AI: An AILit Framework for Primary and Secondary Education. “EU AI Literacy Framework” is an informal name people use because the European Commission co-authors it.
Does the EU AI Act require a specific AI literacy framework?
No. Article 4 of the EU AI Act, which entered application on 2 February 2025, requires providers and deployers of AI systems to ensure “a sufficient level of AI literacy” of their staff. It does not prescribe a framework or require formal certification. The European Commission’s AI Literacy Q&A notes that simply asking staff to read an AI system’s instructions for use is generally not sufficient.
What’s the difference between AI literacy and AI competency?
In most frameworks, literacy refers to understanding (you can explain what AI is, where it fails, what its risks are), and competency refers to applied ability (you can use AI critically and effectively in practice). UNESCO uses “competencies” for both. ETS treats the two as a progression. Chiu and colleagues (2024) make the distinction explicit in their Hong Kong-developed framework. The vocabulary matters less than the structure underneath it.
Which framework is best for higher education specifically?
None of the five frameworks above are higher-education-specific. For universities, the more directly applicable frameworks are from EDUCAUSE, WCET, Digital Promise, and the Digital Education Council. UNESCO’s teacher framework adapts reasonably well to faculty CPD, but the underlying assumption is teacher-in-classroom rather than lecturer-and-seminar.
How often do these frameworks update?
UNESCO (2024) and AI4K12 (2019, ongoing) are stable. The OECD-EC AILit framework is currently in draft (May 2025), with the final version scheduled for 2026. The ETS research report is from 2025 and has not yet been revised. Expect a refresh cycle of every two to three years across the field, with more frequent updates likely as the EU AI Act exemplars are released.
Can we adopt more than one framework?
Yes, and most institutions that take this seriously do. The frameworks aren’t mutually exclusive. A common pairing is AI4K12 for K-12 curriculum content plus UNESCO’s teacher framework for staff CPD. Add OECD-EC AILit if you operate in the EU. The frameworks were designed by different groups for different jobs; layering them is closer to how the field actually expects them to be used than picking one and ignoring the rest.