Status: 6 September 2026

AI in Europe's Schools – The Transition to the Anchoring Phase

Analysis of 152 official primary documents from 49 countries and regions. Europe is transitioning from the recommendation phase (2023/24) to the anchoring phase (2025/26).

152
Documents analysed
49
Countries & Regions
26%
Mandatory curricula
84
Documents 2024–26
Download Full Report (PDF)

Overall Landscape & Temporal Dynamics

European AI education policy has evolved in three waves: early digital strategies, the GenAI surge from 2022, and the current anchoring phase with legal acts.

Documents by Year of Publication
Three waves of AI education policy
Key Finding
84 documents were produced in 2024–2026 – more than twice as many as in the previous eight years combined. The third wave is characterised by statutory regulations and curriculum amendments rather than mere guidelines.
Document Types at a Glance
From recommendations to mandatory curricula
Important Note
45% of all documents are recommendations or reference frameworks. Yet 39 documents (26%) already have mandatory status – remarkable for such a young topic.
The Three Waves of AI Education Policy
2013–2021: First Wave
Digital strategies and computing curricula. AI appears as a sub-aspect of media literacy (e.g. English National Curriculum Computing 2013, Swiss Lehrplan 21, KMK Strategy 2016).
2022–2024: Second Wave (GenAI Surge)
Rapid guidelines on dealing with ChatGPT & Co. – mostly without binding character. Examples: KMK Recommendation 2024, NRW Action Guide, UNESCO Guidance 2023.
2025–2026: Third Wave (Anchoring)
Statutory regulations, curriculum amendments and binding framework curricula. Examples: Slovakia Dodatok č.17, Poland Podstawa programowa, Sweden Gy25, Italy Indicazioni 2025.

Binding Force & Governance

Binding force correlates with curricular integration, not with the density of guidelines. The binding ladder shows seven levels of governance.

The Binding Ladder
Seven levels of descending binding force
1. Supranational Law (AI Act, Council of Europe) 2
2. National Legal Acts (Regulations, Official Gazette) 7
3. Mandatory Curricula without Statutory Form 30
4. Electives & Voluntary Offers 6
5. Strategies & Programmes 15
6. Recommendations & Guidelines 69
7. Drafts & Consultations 2
Governance Beyond the Curriculum
Greece (prohibition catalogue), Denmark (exam ban) and England (Product Safety Expectations) show that effective governance also arises from usage rules, exam law and product requirements.
Binding Force by Region
Mandatory quota per geographic cluster
DACH Leads in Mandatory Quota
The DACH cluster has the highest mandatory quota at 40% – however, this is due to the cumulative effect of three federal systems, not central steering.

Country & Regional Comparison

Six geographic-political clusters with different governance modes: curriculum anchors, guideline strategists and federal patchworks.

DACH
DE, AT, CH
25

Curriculum anchoring via computing/digital subjects, flanked by state- or canton-level guidelines.

Western Europe
FR, BE, NL, LU
28

Central-state digital frameworks with AI modules, reference frameworks and certification (Pix, Kerndoelen).

Nordics / Baltics
SE, FI, DK, NO, IS, EE, LV, LT
24

Guideline and training governance with high municipal school autonomy. Sweden and Iceland as pioneers.

Southern Europe
ES, PT, IT, GR, MT, CY
22

Legislative minimum standards plus ministerial guidelines. Portugal as a special case (only recommendations).

Central & Eastern Europe
PL, CZ, SK, HU, RO, BG, HR, SI
25

Youngest curriculum reform waves with explicit AI literacy. Slovakia and Poland as pioneers.

UK / Ireland
ENG, SCT, WLS, NIR, IE
15

Guidance regime of decentralised education nations. Wales with binding Digital Competence Framework.

Typology of Anchoring Modes
Type Characteristics Example Countries
Curriculum Anchors AI as explicit mandatory content in curricula or legal acts; student competency goals at the centre PL, SK, IT, SE, FR, GR, HR, ES
Guideline Strategists Ministerial recommendations and training programmes as main instrument; no direct curriculum change DK, FI, AT, IE, NL
Federal Patchworks Competence shifted to state/regional level; heterogeneous mix of mandatory subjects and guidelines DE, CH, BE, UK, ES

Target Group Analysis

Teachers are the most strongly addressed target group (88 documents), followed by lower secondary education. Vocational education remains the largest gap.

Addressed Target Groups
Multiple mentions per document possible (n=152)
Teachers in Focus
Almost exclusively in-service training and continuing education for existing teachers. Initial teacher education and induction remain largely unaddressed – the lever with the longest-term impact.
Structural Gaps
The biggest coverage gaps in the corpus
Vocational Education (TVET) 10 / 152
Initial Teacher Education & Induction barely addressed
Upper Secondary (ISCED 3) 45 / 152
Primary Education 48 / 152
Lower Secondary (ISCED 2) 72 / 152
Teachers (In-service Training) 88 / 152
TVET – The Biggest Gap
Only 10 of 152 documents explicitly address TVET, even though AI-workplace references appear as learning content in 14 student documents. Spain (RD 279/2021) and Denmark are pioneers here.

Competency Priorities

Two clearly separated profiles: students are to understand, apply and assess AI. Teachers are to take responsibility for, shape and safeguard AI use.

Competency Topics: Students vs. Teachers
Distribution across 84 student and 88 teacher-related documents
Complementary Profiles
Ethics is the only continuous consensus (36 vs. 56 mentions). Technical depth (ML/programming) is a student domain (37 vs. 6), while didactics and wellbeing are teacher-dominated. Critical thinking is the most balanced topic at 31:30.
Student Profile
"Understand, apply, assess"
Teacher Profile
"Take responsibility, shape, safeguard"
Reference Frameworks
Normative architecture of AI competency
DigComp 2.2 (EU)
Annex: Citizens interacting with AI systems
UNESCO AICFS / AICFT
Competency frameworks for students & teachers
AILit Framework (EU/OECD)
4 domains, 23 competencies – basis for PISA 2029
EU Ethical Guidelines
Human dignity, fairness, transparency, privacy

Integration Models & Didactics

Three integration forms exist in the tension between breadth and depth. Didactically, all systems converge on a three-mode schema.

1 Stand-alone Subject
Highest depth, limited reach
  • Sweden: "Artificiell intelligens" (Gy25) – 2 levels à 100 points
  • France: NSI (Speciality) + SNT (Mandatory)
  • Luxembourg: "digital sciences" (Mandatory ISCED 2)
  • Ireland: Leaving Certificate Computer Science
Advantage: Maximum content depth, clear accountability, graded.
Disadvantage: Selective reach, teacher shortage, late introduction.
2 Cross-curricular Theme
Maximum breadth, diffuse responsibility
  • Slovakia: "AI gramotnosť" as cross-curricular theme (from 2026)
  • Cyprus: Horizontal integration across all subjects
  • Wales: Digital Competence Framework (DCF)
  • Finland: AI as extended literacy
Advantage: All students, all levels, subject context.
Disadvantage: Low guaranteed depth, implementation depends on each teacher.
3 Module in Computing
The most common middle way
  • Poland: Informatyka from 2026 (AI tools, hallucinations)
  • NRW (DE): "Automata and AI" (Grades 5/6)
  • Bavaria (DE): LehrplanPLUS AI (Grade 11)
  • Greece: Dedicated AI unit in Gymnasio
Advantage: Appropriate placement, uses existing subject structures.
Disadvantage: Reach only as large as the computing subject itself.
Didactic Convergence: The Three-Mode Schema
Regardless of integration form, all systems draw on identical basic schemata
🎯
Learning ABOUT AI
Functioning, limits, ethics, bias, hallucinations
Cyprus: "Learning about AI"
Spain: "enseñar sobre la IA"
🔧
Learning WITH AI
Tool use, prompting, AI-supported projects
Luxembourg: "Apprendre avec l'IA"
Denmark: "undervisning med genAI"
🛡️
Learning FOR AI
Competencies AI cannot replace: criticism, creativity
Romania: "Teaching FOR AI"
Ireland: 4Cs (Critical Thinking, Creativity)

Conclusion & Recommendations

Five strategic recommendations for education policy, school practice, teacher education, vocational education and the European level.

01
Choose the Combination Model
None of the three pure integration forms solves the breadth-depth dilemma. The most robust model is the combination of a mandatory module for depth and a cross-curricular element for breadth – realised by Poland (Informatyka + Kompas Jutra), Slovakia (Dodatok č.17 + graded subject addenda) and France (SNT/NSI + Pix pathway).
02
Clarify Rules Before Tool Rollout
The most effective governance instruments lie beyond the curriculum. Greece (prohibition catalogue), Denmark (exam ban) and Ireland (disclosure obligation) clarified the rules of use before introducing tools – not after. Governance is a school leadership task.
03
Address All Three Training Phases
Teachers are the most strongly addressed target group, but the address is concentrated almost entirely on in-service training. Initial teacher education and induction remain unaddressed – the lever with the longest-term impact. Sweden's 15-ECTS coupling shows what such a mechanism can look like.
04
Scale Up TVET Specialisations
With only 10 of 152 documents, TVET is the thinnest-covered target group. Spain (RD 279/2021) and Denmark (STUK recommendations) provide the robust counter-model. Occupation-specific AI specialisations must be scaled, otherwise a cohort will emerge that knows AI as literacy but not as vocational competency.
05
Mirror National Frameworks to AILit
Three supranational drivers create binding adaptation pressure: AI Act Art. 4 (applicable since Feb. 2025), the AILit Framework (4 domains, 23 competencies) and PISA 2029 (Media & AI Literacy). National competency frameworks should mirror their domain structure to AILit to ensure compatibility with the assessment and EU law.

Overall Conclusion

European AI education policy in 2026 is characterised by a distinctive asymmetry: on the content-didactic level there is far-reaching convergence – ethics and data protection, critical output assessment, human oversight, the three-mode schema and the prohibition of fully automated assessment form a shared minimum core. On the structural level, however, the gaps remain considerable: vocational education, initial teacher training, primary education in Southern Europe, binding evaluation and the technical qualification of teachers are the blind spots of the anchoring phase.

The decisive test in the coming years will therefore be less the invention of new instruments than the closing of structural gaps – and the question of whether the pending enactments from 2026/27 will exert the expected pull effect on the still undecided systems.

Development Paths to 2029
2026/27 – Implementation Phase
Enactment in Poland (Sept. 2026), Italy, Netherlands, Romania, Slovakia (upper secondary). First evaluations of the anchoring phase.
2027 – AI Act High-Risk
Educational institutions as operators become addressees of concrete compliance obligations (high-risk classification of education-related AI applications, Annex III).
2029 – PISA & New Reforms
PISA 2029 assesses "Media & AI Literacy" internationally for the first time. Slovenia's curriculum reform KUP (target horizon 2029), Czech RVP revision.
Download Full Report (PDF)