⚡ Quick Summary
This guide is written by the educator and school leader Matthew Wemyss and sets out what the EU AI Act means for schools. It describes itself as a practical starting point rather than legal advice, states that it was reviewed by the lawyer Dr Marta Claudia Cliza, and reflects the Act as of March 2025. The book runs in three parts, moving from why compliance matters to assessment and review to AI literacy.
Part 1 explains that education is treated as high-risk under the Act and works through the four risk levels — unacceptable, high, limited and minimal — quoting Annex III on education and vocational training, and setting out fine tiers of €35 million or 7% of global turnover for prohibited practices, €15 million or 3% for high-risk violations and €7.5 million or 1% for misleading information. Part 2 covers mapping AI use, reviewing tools and building compliance systems. Part 3 turns to AI literacy, UNESCO's competency framework, staff INSET and safe testing environments. The guide ends with a resources pack, a summary table of what the Act means for schools and a self-assessment scorecard with action steps.
🧩 What’s Covered
- Scope and limits: what the guide is (a practical starting point, an unpacking of the Act as of March 2025, a governance tool) and is not (legal advice, a guarantee of compliance, a technical manual, one-size-fits-all), plus considerations including AI regulatory sandboxes, EU database registration, whistleblower protection and accessibility.
- Keywords and timeline: definitions of AI system, deployers, providers, the four risk levels, GPAI, conformity assessment, CE marking, FRIA, human oversight and post-market monitoring, and the bodies involved (AI Office, European AI Board, national competent authorities); dates run from August 2024 entry into force through February 2025 prohibitions and AI literacy, May 2025 GPAI codes of practice and August 2026 to August 2027 and December 2030.
- Risk scenarios: six worked examples — AI grading, over-reliance on adaptive platforms, admissions and career guidance, a data breach, emotion recognition and AI-driven learning paths — each with the risk and the Act's response, including the ban on biometric emotion recognition in education.
- Risk levels and GPAI: tabulates unacceptable, high, limited and minimal risk with school examples, quotes Annex III on education and vocational training, covers exceptions for narrow, supportive or purely analytical tools, and explains GPAI with and without systemic risk, including the 10^25 floating point operations threshold.
- Penalties and deployer duties: the three fine tiers, the factors that shape penalties (severity, intent, remedial action, repeats, size, financial gain) and schools' position as deployers rather than developers.
- Self-assessment scorecard: nine statements rated 1–5 with bands from AI Beginner to AI Pioneer, an action-setting step and a Part 1 recap.
- Assess: why assessment comes first, the problem of "shadow AI", and steps to map AI use, survey tools, assess AI literacy, classify risk, check data compliance and identify gaps.
💡 Why it matters?
For school leaders, data protection officers, IT teams and compliance advisers, the guide turns a long, technical regulation into decisions a school can actually take: which tools to inventory, which to treat as high-risk, what human oversight means for grading or admissions, and where GDPR and copyright duties overlap with the Act. It connects the Act's obligations to everyday realities such as vendor contracts, free tools, wrapper apps and APIs. Because it is explicit that it is not legal advice and that national enforcement may add requirements, it works as a scoping document to use before consulting regulators or lawyers.
❓ What’s Missing
The guide states its own limits: it is not legal advice, not a guarantee of compliance, not a technical manual and not one-size-fits-all, and it can only reflect the Act as of March 2025 while implementation guidelines and national enforcement are still to come. It leaves unresolved the tension between GDPR's restrictions on sensitive data and the Act's expectation that bias be tested, notes that career guidance is not explicitly covered by the Act, and includes a speculative aside about robots in design and technology workshops. This review covers only pages 1–80 of 258, so the chapters on review, compliance, AI literacy and the resources pack were not examined.
👥 Best For
Best for school and college leaders, designated AI leads and data protection officers who need to understand the EU AI Act's reach into education and to start an inventory and risk classification. It also suits IT teams assessing vendors, wrappers and APIs, and legal or compliance advisers wanting a plain-language briefing on the Act's education provisions.
📄 Source Details
AI in Education: An EU AI Act Guide by Matthew Wemyss. No publisher, imprint or publication date is printed; the text states it reflects the EU AI Act as of March 2025 and that the author's AI in Education Handbook series is available on Amazon. The supplied PDF runs to 258 pages, with contents listing chapters, a Resources Pack (page 245) and an Appendix summary table (page 246). The work is released under a Creative Commons Attribution 4.0 International Licence. The input was a partial text extraction covering pages 1–80 of 258; sections beyond chapter 6 were not available.
