evidoria

← Back to browse

Good practice

AI Toetsingskader — SIVON and Kennisnet's EU AI Act Compliance Framework for Dutch Schools

Netherlands · Zoetermeer · See the Netherlands profile

SIVON and Kennisnet, backed by the Dutch education ministry, published a step-by-step AI Toetsingskader (v1.0, April 2026) helping the Netherlands' 476 school boards and their suppliers classify AI tools under the EU AI Act and identify which need extra safeguards.

476 primary and secondary school boards
School boards covered by SIVON's guidance (as of April 2026)

Details

Promoter
SIVON / Kennisnet (with the Dutch Ministry of Education, Culture and Science)
Period
{'end': '', 'start': '2026-04'}
Keywords
AI governance, EU AI Act compliance, data protection, school IT procurement, primary and secondary education

Context

SIVON, the ICT cooperative representing 476 primary- and secondary-school boards in the Netherlands, publishes procurement and compliance guidance on behalf of Dutch schools. As generative AI tools spread through classrooms and school administration and the EU AI Act's obligations began phasing in from August 2024, schools and their software suppliers lacked a shared way to work out which rules applied to which tool. On 1 April 2026, SIVON published version 1.0 of the AI Toetsingskader Funderend Onderwijs (AI Assessment Framework for Primary and Secondary Education), developed with Kennisnet and supported by the Dutch Ministry of Education, Culture and Science (OCW), explicitly marked as preliminary with updates planned through 2026.

Objectives

To give schools and their AI suppliers a shared, step-by-step way to classify AI tools under the EU AI Act — whether a system meets the Act's definition of an AI system, whether it falls under a prohibited practice, which risk category applies, and what transparency or exemption provisions follow.

Activities

The framework flags admissions decisions, student-performance evaluation and exam monitoring as likely high-risk uses, and warns schools that customising a generative AI tool for their own teaching use can itself make the school an AI 'provider' under the Act, taking on full supplier compliance duties rather than just a deployer's. It sits alongside SIVON's own Data Protection Impact Assessment (DPIA) of Google Gemini within Workspace for Education (results due in 2026), following sister organisation SURF's 2024 DPIA of Microsoft 365 Copilot, which concluded that cautious ('terughoudend') deployment was still warranted. Related SIVON guidance tells schools to prefer paid accounts that exclude submitted data from model training, to manage AI accounts centrally rather than let staff install tools unsupervised, and not to let students under 13 use generative AI directly.

Conclusions

The framework is a compliance and risk-classification tool, not a pedagogical intervention: it contains no data on student learning outcomes, and both SIVON and Kennisnet describe it as a preliminary version they expect to revise as the EU AI Act's secondary guidance is finalised.

Implementation
Indicative cost
Low (< €50k) — No budget figures are published. Guidance-document production by an existing ICT cooperative and its partners, without new technical infrastructure, points to a low direct cost; treated as a conservative estimate.
Time to results
Short (< 1 year) — Version 1.0 was published on 1 April 2026 with updates explicitly planned through the rest of 2026 as EU implementing guidance solidifies — a short, iterative revision cycle rather than a long-horizon programme.
Staffing & skills
SIVON policy and compliance staff, Kennisnet, Support from the Dutch Ministry of Education, Culture and Science (OCW)

Conditions for success

  • Multi-stakeholder development (SIVON, Kennisnet, OCW) giving the framework sector-wide legitimacy
  • Pairing the general framework with concrete, tool-specific work — SIVON's DPIA of Google Gemini in Workspace for Education and SURF's prior DPIA of Microsoft 365 Copilot
  • Explicit, actionable guidance (e.g. prefer paid accounts that exclude data from model training, manage AI accounts centrally, no direct generative-AI use for under-13s)

Common failure modes

  • The framework is explicitly preliminary and expected to be revised as EU AI Act secondary guidance is finalised, so current classifications may change
  • A school that customises a generative AI tool for its own teaching use can inadvertently become an AI 'provider' under the Act, taking on a compliance burden it may not anticipate
  • No data on student learning outcomes exists — this is a compliance tool, not a validated pedagogical or efficiency intervention

Where it fits

Governance type
national school-ICT cooperative with ministry support
Scale
national (476 school boards)
Income level
high income

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful