Catalonia sets out 45 guidelines on which uses of AI to promote and which to limit in schools
The Department of Education and Vocational Training publishes a guide in September 2026 that advises against autonomous use of general-purpose AI by under-14s, grading with AI and AI-text detectors.
The Department of Education and Vocational Training of the Generalitat de Catalunya published in September 2026, as a first edition, the Orientacions per a l’ús de la intel·ligència artificial als centres educatius (guidelines for the use of artificial intelligence in schools), drawn up by its Directorate-General for Curriculum and Vocational Development. It updates the 2024 document La intel·ligència artificial en l’educació. It was written by an internal working group between April and June 2026 and reviewed by a group of experts from the administration, universities, the inspectorate, data protection, school leadership and teaching. It runs to about seventy pages. The text itself states that AI models were used to review drafts and to write the examples in an annex.
The guide starts from the premise that students already use AI to study and do homework. In its words, the question is “not whether AI should enter classrooms or not, but how we want it to enter”. It justifies its caution with five reasons: AI solves a good part of the activities, AI-text detectors do not offer sufficient guarantees (especially outside English), general-purpose AI is not designed for learning, the available evidence is limited but points to risks if it is used without guidance, and there is no European public AI of reference, while US legislation (CLOUD Act, FISA) casts doubt on the real protection of data.
At its core are 45 numbered guidelines (OIA-01 to OIA-45), organised by area: school, teachers, educational services, support staff and families. Students under 14 should not use general-purpose AI autonomously; if it is used in class, the teacher operates it. Between 14 and 18, families’ consent is not needed, because the legal basis is the public-interest mission, but collecting their authorisation is recommended as a safeguard, compulsory when the tool’s terms require it, and direct interaction by students is placed from the second term of the third year of ESO (compulsory secondary education) onwards. It also asks for training and persistent memory to be switched off, agentic AI to be avoided, no personal data to be entered, and each school to approve a framework for use with a register of tools.
The pedagogical part is the most useful for the classroom. It asks teachers to keep the lead, verify what AI generates, declare their own use on a common scale and design activities that require interpretation and a personal position. In assessment, AI must not grade or determine the level of attainment, as this is considered a high-risk use under the European AI Act; it may help to create instruments, mark unambiguous answers or detect common errors using pseudonymised data. On detectors it concludes that they are inadvisable: “the best detectors are still people”.
It should be read with its limits in mind: these are guidelines, not rules, and each school must adapt them; several sections refer to external pages that are not in the PDF; and the text admits that the evidence is limited and the technology changes quickly. What can be transferred outside Catalonia are the criteria, not the procedures. A good starting point is annexes 1 to 3, one-page infographics with general guidelines, for students and for families, and then guidelines OIA-28 and OIA-36 to OIA-38 on pedagogical use and assessment. Annex 4 offers four analogies (basketball, dance, rap, video games) to explain why learning is still worthwhile.
Catalonia sets out 45 guidelines on which uses of AI to promote and which to limit in schools
The Department of Education and Vocational Training publishes a guide in September 2026 that advises against autonomous use of general-purpose AI by under-14s, grading with AI and AI-text detectors.
The Department of Education and Vocational Training of the Generalitat de Catalunya published in September 2026, as a first edition, the Orientacions per a l’ús de la intel·ligència artificial als centres educatius (guidelines for the use of artificial intelligence in schools), drawn up by its Directorate-General for Curriculum and Vocational Development. It updates the 2024 document La intel·ligència artificial en l’educació. It was written by an internal working group between April and June 2026 and reviewed by a group of experts from the administration, universities, the inspectorate, data protection, school leadership and teaching. It runs to about seventy pages. The text itself states that AI models were used to review drafts and to write the examples in an annex.
The guide starts from the premise that students already use AI to study and do homework. In its words, the question is “not whether AI should enter classrooms or not, but how we want it to enter”. It justifies its caution with five reasons: AI solves a good part of the activities, AI-text detectors do not offer sufficient guarantees (especially outside English), general-purpose AI is not designed for learning, the available evidence is limited but points to risks if it is used without guidance, and there is no European public AI of reference, while US legislation (CLOUD Act, FISA) casts doubt on the real protection of data.
At its core are 45 numbered guidelines (OIA-01 to OIA-45), organised by area: school, teachers, educational services, support staff and families. Students under 14 should not use general-purpose AI autonomously; if it is used in class, the teacher operates it. Between 14 and 18, families’ consent is not needed, because the legal basis is the public-interest mission, but collecting their authorisation is recommended as a safeguard, compulsory when the tool’s terms require it, and direct interaction by students is placed from the second term of the third year of ESO (compulsory secondary education) onwards. It also asks for training and persistent memory to be switched off, agentic AI to be avoided, no personal data to be entered, and each school to approve a framework for use with a register of tools.
The pedagogical part is the most useful for the classroom. It asks teachers to keep the lead, verify what AI generates, declare their own use on a common scale and design activities that require interpretation and a personal position. In assessment, AI must not grade or determine the level of attainment, as this is considered a high-risk use under the European AI Act; it may help to create instruments, mark unambiguous answers or detect common errors using pseudonymised data. On detectors it concludes that they are inadvisable: “the best detectors are still people”.
It should be read with its limits in mind: these are guidelines, not rules, and each school must adapt them; several sections refer to external pages that are not in the PDF; and the text admits that the evidence is limited and the technology changes quickly. What can be transferred outside Catalonia are the criteria, not the procedures. A good starting point is annexes 1 to 3, one-page infographics with general guidelines, for students and for families, and then guidelines OIA-28 and OIA-36 to OIA-38 on pedagogical use and assessment. Annex 4 offers four analogies (basketball, dance, rap, video games) to explain why learning is still worthwhile.
