Allowed: trigger a review
With reasonable suspicion, you may initiate an additional check - a conversation, an oral follow-up on content, a second opinion via a tool. That's the normal case, not a special power.
!
In short
A detector score alone is not legal proof of misconduct - it triggers a review, it doesn't replace one. When there's a reasonable suspicion, you may hold a clarifying conversation, get a second opinion, and act within your school's or institution's examination policy. What you may not do: change a grade or record misconduct based on a percentage alone, without giving the student a chance to respond.
The legal framework, briefly
Three layers interact: your state's school law (sets the general rules for handling suspected misconduct), your institution's examination regulations (set the concrete process - deadlines, the right to be heard, appeals), and since 2 August 2026 the EU AI Act labelling duty (Art. 50, aimed primarily at providers of AI content, but increasingly shaping what schools and examination policies expect in their own rules). None of these three layers turns an automated score into proof.
WHAT YOU CAN ACTUALLY DO
What educators may do with an AI-suspected text
Allowed: run a digital check
Processing a submitted text through an AI detection tool is permitted when the institution discloses it transparently and has a data-processing agreement with the provider.
Not allowed: change a grade without a hearing
A decision with consequences (grade deduction, a misconduct record) requires the right to be heard - the student must get a chance to respond to the allegation before the decision is final.
Not allowed: rely on the score alone
A single percentage with no further reasoning rarely survives scrutiny - especially for non-native-speaking students, where the error rate is structurally higher.
🛈
Practical tip: decide the process in advance, not case by case. A school or examination policy that clearly covers AI use, the review process on suspicion, and the data-protection basis protects both you and your students - and makes every individual decision easier to defend.
At a glance: practical recommendations for educators
| Topic | Legal basis & provisions | Practical consequence |
|---|---|---|
| Evidentiary value of AI scoresAn AI score alone is not proof of misconduct - it only triggers a review. | Detectors don't provide full proof (§ 24 VwVfG, the official-investigation principle). Proving misconduct requires a free evaluation of evidence plus further indicators. | Never use an AI percentage as the sole reason for sanctions; always gather solid indicators (e.g. hallucinations, logic gaps, comparison with the student's other texts). |
| Power to investigateWith reasonable suspicion, further review steps may be initiated. | An initial suspicion authorises the school to investigate under its pedagogical discretion. | Ask content-related questions about the text, hold a clarifying subject discussion, or get a second opinion. |
| Data protection & softwareDigitally scanning texts is allowed when it's transparent and a data-processing agreement (DPA) is in place. | Student texts are personal data (Art. 4 GDPR). Processing requires a legal basis plus a data-processing agreement (Art. 6, 13 & 28 GDPR). | Only use tools centrally licensed by the school with a DPA in place; never use private or unlicensed detectors. |
| Due process & the right to be heardGrade deductions or misconduct records may not be issued without a hearing. | Obligation to grant the right to be heard (§ 28 VwVfG, Art. 103(1) Basic Law analogously) to safeguard Art. 12 Basic Law. | Hear the student out before the final decision on the suspicion and document their statement. |
| Non-native speakersAI scores are unreliable for non-native speakers due to structurally higher false-positive rates. | Typical indicators can be undermined by atypical patterns of events or language use. | Account for language level; use drafts, notes and version history from the student to clarify the case. |
| EU AI Act & guidelinesEU law shapes the framework for school and examination policies. | Transparency duties under Art. 50 EU AI Act; automated assessment systems count as high-risk AI (Annex III no. 3(b)). | Avoid fully automated decisions; use AI detectors only as an aid under human oversight. |
Frequently asked questions
Usually not. Most examination regulations require additional evidence and a right to be heard - the score triggers a review, it doesn't automatically decide the outcome.
Yes, transparency is a GDPR requirement whenever personal data (the submitted text) is processed. Easiest to cover in the school or examination policy, not case by case.
Article 50 is aimed primarily at providers and distributors of AI content in commercial/public contexts, not directly at student assignments. Still, it shapes public expectations - schools should have their own clear rules rather than relying on it alone.
By Simon Schmidlin
Keywordsai in the classroom law, eu ai act school, ai labelling requirement school, legal basis ai suspicion student, school law ai use