Governments, regulators and international bodies have published a great deal of official guidance telling schools what to do about AI. This page reads it at source and asks one question of each document: did it measure anything, or does it assert a position and recommend a practice? Of the 13 documents read here, 3 present any original data at all. The other 10 are guidance, frameworks or law.
That is not a complaint. Guidance is supposed to be guidance, and a regulator writing an assessment rule is doing its job whether or not it ran a trial. It matters because the volume of official documents is routinely mistaken for a weight of evidence, and the issuing bodies are more candid about this than the people citing them. The UK Department for Education says so on the face of its own policy.
We have limited evidence on the impact of AI use in education on learners’ development, the relationship of AI use and educational outcomes, and the safety implications of children and young people using this technology in the classroom.
UK Department for Education, generative AI in education, updated 12 August 2025
The three that measured something
These carry original data, and the data is thinner than the documents around it suggest.
- UNESCO, Guidance for generative AI in education and research. A UNESCO survey of more than 450 schools and universities found fewer than 10 per cent had any formal institutional guidance on generative AI as of mid-2023. That single figure is the measured part; the rest is prescription. Read at source.
- UNESCO, AI competency framework for teachers. A count rather than a study: as of 2022, seven countries had an AI competency framework or professional-development programme for teachers. Read at source.
- UNICEF Innocenti, Guidance on AI and children: recommendations for AI policies and systems that uphold child rights, Version 3.0. Grounded in an original UNICEF study across twelve countries with children and caregivers, which makes this evidence-informed guidance rather than assertion. The document itself remains a set of recommendations. Read at source.
What each document actually says
Grouped by who issued it. Every entry was fetched and read on the issuing body’s own domain, and carries a verbatim line so the reading can be checked. The label states what the document is: guidance carries no binding force, a framework is a structure for others to adopt, and binding means it has legal or regulatory effect.
International bodies
Guidance for generative AI in education and research · UNESCO · 7 September 2023
Guidance
Seven steps for governments: national data-protection and privacy standards, human-rights-based approval processes before a tool reaches a classroom, and mandatory teacher training. Sets an age floor of 13 for classroom use of AI tools without parental consent, and asks institutions to validate tools before approving them.
Generative AI can be a tremendous opportunity for human development, but it can also cause harm and prejudice. It cannot be integrated into education without public engagement, and the necessary safeguards and regulations from governments.
Audrey Azoulay, UNESCO Director-General
AI competency framework for students · UNESCO · 2024
Framework No original data.
Twelve competencies across four dimensions, at three progression levels, for insertion into national curricula so that students become users and co-creators of AI rather than consumers of it.
Integrating AI learning objectives into official school curricula is crucial for students globally to engage safely and meaningfully with AI.
AI competency framework for teachers · UNESCO · 2024
Framework
Fifteen competencies across five dimensions, at three levels, as the reference for national teacher-training and assessment programmes.
As of 2022, however, only seven countries had developed an AI competency framework or professional development programme for teachers.
Guidance on AI and children: recommendations for AI policies and systems that uphold child rights, Version 3.0 · UNICEF Innocenti · December 2025
Guidance
Ten requirements and 48 recommendations. Requirement nine covers education: update AI-literacy curricula, train and equip teachers so that AI is placed around them rather than in place of them, and adopt AI in education only where it is appropriate and evidence-based. Version 3.0 adds AI companions, AI-generated child sexual abuse material, and child labour in the AI supply chain.
AI has transformed the traditional teacher-student relationship into a teacher-AI-student dynamic.
United Kingdom
Generative artificial intelligence (AI) in education · UK Department for Education · Published 26 October 2023, last updated 12 August 2025
Guidance No original data.
Leaves adoption to schools and colleges while requiring compliance with data protection, safeguarding and copyright law. Advises that personal data is not entered into generative tools, notes that schools report clearer benefits from teacher-facing use than pupil-facing use, and defers to Ofqual and JCQ on assessment integrity.
We have limited evidence on the impact of AI use in education on learners' development, the relationship of AI use and educational outcomes, and the safety implications of children and young people using this technology in the classroom.
Artificial intelligence malpractice and assessment: advice note · Ofqual · 27 April 2026
Guidance No original data.
Awarding organisations should assess each assessment's vulnerability to AI malpractice by its features: task specificity, output type, supervision level and stakes. It warns that the obvious mitigations, such as removing internet access, can damage the validity of the assessment they protect.
In some cases, AOs may conclude that there are no effective or proportionate measures available to adequately address the risk of AI-related malpractice for a particular assessment.
AI Use in Assessments: your role in protecting the integrity of qualifications · Joint Council for Qualifications · Published 26 April 2023, revision two 30 April 2025
Binding No original data.
Teachers may accept only work that is the student's own. Misuse of AI such that submitted work is not the student's own is malpractice and carries severe sanctions. Students must identify any AI-generated passage reproduced directly.
Students who misuse AI to the extent that the work they submit for assessment is not their own will have committed malpractice in accordance with JCQ regulations and could attract severe sanctions.
European Union
Artificial Intelligence Act, Annex III paragraph 3: education and vocational training · European Union · Regulation (EU) 2024/1689, dated 13 June 2024; high-risk obligations apply from 2 August 2026
Binding No original data.
Classifies as high-risk any AI system used to determine admission to educational institutions at any level, to evaluate learning outcomes, to assess the level of education a person should receive, or to monitor and detect prohibited behaviour during tests. High-risk classification triggers the full Chapter III obligations: risk management, data governance, technical documentation, human oversight and conformity assessment.
AI systems intended to be used to evaluate learning outcomes, including when those outcomes are used to steer the learning process of natural persons in educational and vocational training institutions at all levels.
United States, federal
Artificial Intelligence and the Future of Teaching and Learning: insights and recommendations · US Department of Education, Office of Educational Technology · May 2023
Guidance No original data.
Seven recommendations, including keeping humans in the loop, aligning models to a shared vision for education, and requiring teacher-facing AI to be inspectable, explainable and overridable. Frames itself around open questions rather than settled answers.
Always Center Educators in Instructional Loops.
United States, states
Guidance for the Safe and Effective Use of Artificial Intelligence in California Public Schools · California Department of Education · Updating the 2023 Learning With AI, Learning About AI guidance
Guidance No original data.
Human-centred AI use, AI literacy, equitable access, academic integrity and data privacy. States on its own face that compliance is not mandatory, under Education Code section 33308.5.
compliance with this guidance is not mandatory
Guidance on the use of artificial intelligence in schools · North Carolina Department of Public Instruction · 16 January 2024
Guidance No original data.
Promotes the EVERY framework for classroom use: evaluate, verify, edit, revise, and you are responsible. Presses for AI literacy across all grades and subjects. North Carolina was the fourth US state to publish such guidance.
Evaluate, Verify, Edit, Revise, You're responsible
AI Model Policy for Ohio Districts and Schools · Ohio Department of Education and Workforce · Published by 31 December 2025, districts must adopt by 1 July 2026
Binding No original data.
Issued under House Bill 96 and Ohio Revised Code section 3301.24, which requires the department to publish a model policy and requires every district, community and STEM school to adopt a local AI policy by 1 July 2026. The only US state entry in this corpus with statutory force rather than advisory status.
AI Model Policy for Ohio Districts and Schools
Australia
Australian Framework for Generative Artificial Intelligence in Schools · Australian Government Department of Education · Approved 5 October 2023, in force from Term 1 2024, review endorsed June 2025
Framework No original data.
Tools used in schools must uphold privacy and data rights, comply with Australian law, avoid unnecessary collection of student data, limit retention, prevent onward distribution and prohibit its sale. Reviewed at least annually.
schools should not use generative AI products that sell student data.
Jason Clare, Minister for Education
What could not be confirmed
UAE Ministry of Education, Compulsory AI subject, kindergarten to grade 12. Reported by Gulf News, The National, Khaleej Times, Arab News and Emirates 24|7 as announced on 4 May 2025 for the 2025-2026 academic year, taught inside the existing Computing, Creative Design and Innovation course and assessed by project work. No document on moe.gov.ae, the UAE Cabinet site or WAM could be retrieved carrying the announcement itself. Two genuine Ministry pages on other AI programmes were read, so the domain is reachable; the announcement text is not. The consistency of the press coverage is not a substitute, because all five outlets appear to derive from the same social-media post.
It stays out of the corpus until a document can be read. The reason for saying so rather than leaving a gap is that this is exactly how an unverified claim enters a reference: several reputable outlets carry it, the agreement between them reads as corroboration, and the common source underneath is never checked.
How to use this corpus
Cite guidance for what it is. A UNESCO framework establishes what an international body recommends, and it establishes nothing about whether the recommendation works. If you need to know whether AI helps or harms learning, the measured findings are in the graded evidence base, where Bastani’s school trial and the learning-science work on retrieval practice and productive struggle sit with their limits stated.
The gap between the two is the useful finding. There is a large and growing body of official instruction about classroom AI, and a small body of measurement underneath it, and the documents themselves are clearer about that than their readers usually are.
Related SuperSkills research
On what the evidence does show: how humans learn with AI, retrieval practice, productive struggle and desirable difficulty. On assessment: assessing students when AI can do the assignment and does AI detection work. On what accumulates when practice stops: capability debt. For families: should children use AI and how much should teenagers use AI. The wider reading list is at the AI reports worth reading.
About this research
Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. He has worked in and around education technology for a decade, and was writing weekly on AI and education from 2017, including Reinventing Education (5 November 2017), AI in Education (14 October 2018), AI Grading (16 August 2020) and AI Tutors (13 September 2020). Those pieces were curating and flagging the question rather than advancing a thesis, which is the honest description of them. Every document on this page was read at the issuing body’s own site on 30 August 2026, and anything that could not be is listed above rather than summarised.
Cite this
Hirji, R. (2026). The official guidance on AI in education. The SuperSkills Intelligence Company. Last reviewed 30 August 2026. thesuperskills.com/research/official-guidance-on-ai-in-education