Educational institutions have a characteristic that few other sectors share: the users of AI are, for a large part, the people the institution is meant to educate, not to control. Students and pupils use language models to complete assignments, summaries and papers, often faster and more widely than teachers and policymakers can keep track of. Teachers use those same tools to speed up grading, develop teaching materials or formulate feedback. In between stands a board responsible for educational quality, examinations and the protection of personal data of pupils who are often minors, without always having insight into what happens in practice.
That relationship — use that starts with the user, not with the organization — makes governance in education different from a business environment where IT purchases and rolls out a tool. Here, use arises first, spread across thousands of individual choices, and the question of control only comes up afterwards.
An inventory of educational software shows which digital learning environments, testing systems and administrative packages have been purchased. That list says little about what teachers actually use to create teaching materials, and even less about what pupils and students use to complete assignments. A chatbot that is freely accessible via a browser requires no approval from the IT department and appears in no contract overview.
That makes shadow AI in education a structural fact, not an exception. Use is spread across teachers who individually decide how they deploy a tool, across students who experiment without anyone asking, and across parts of the organization — from scheduling to study advice — that search for tools independently of each other. A complete picture therefore requires not only a technical inventory, but also a question to the people who work with it daily: what do you use, and for what.
That question only yields an honest answer if there is no sanction attached to it. A teacher who admits that assignments are graded with AI assistance, or a student who explains how a paper was put together, will not do so if there is a risk that this use will be treated as a violation. Institutions that primarily want to know what is happening therefore benefit from an approach that first maps out the use and asks the question of desirability and rules separately and later. Those who ask both questions at the same time will get no answer to the first.
Not every use of AI in a school or institution carries the same weight. A tool that helps teachers formulate feedback on an essay has a different impact than a system that counts toward the assessment of an exam, or a tool used in the selection of students for a program. Where AI factors into a decision about an individual pupil or student — a grade, an advice, an admission — the risk level is higher than where AI merely saves time on preparatory work.
That classification by role and risk is needed to properly allocate scarce attention. Not every use requires the same amount of oversight, and a governance approach that treats everything the same loses oversight precisely in the places where the risk is highest.
Educational institutions often already have forms of oversight: an examination board, a participation council, a data protection officer, a quality assurance system. AI governance works better when it connects to these existing structures than when a separate, new circuit is set up alongside them. The examination board that already determines what aids are permitted during a test is the logical place to also assess AI use. The governance question is therefore not only which rules will apply, but also to whom within the existing organization oversight is assigned.
The current legal frameworks — what is mandatory, for whom, and on what timeline — are covered elsewhere. Here, the focus is on the mechanism: knowing what is being used, knowing who is responsible for it, and being able to demonstrate this to the board, the regulator or parents.
The education sector shares with other environments the fact that shadow AI is the norm rather than the exception, but the reason differs. Where AI governance in financial services is mainly driven by oversight of individual customer decisions, and AI governance in the IT sector revolves around AI itself becoming part of the delivered product, in education it concerns a large group of users who hold no formal position within the organization but nevertheless represent the largest share of actual use. That is a different challenge than in AI governance in retail, where use generally does lie with employees in a defined role.
An inventory of which AI is used and by whom is a first step. The question that follows is more specific: which part of the work that teachers, education support staff or policy staff currently do is suitable to be taken over by AI, and which part is not. That is a question per task, not per role or department. The work scan by FTE TO AI calculates this at task level and shows which part of the work can actually be automated, independent of the question of which tools are already circulating for that purpose.
Vraag maar. Governance begint bij weten wat er draait — ook wat niemand heeft goedgekeurd.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.