White Paper November 2026
From detection to verification: assessment towards 2030
Why detecting and banning AI will not hold, and how institutions can move step by step towards verifying what students understand.
This is relevant to you because:
- You are weighing AI detection, process tracking or watermarking and need to know what each can and cannot prove in a misconduct case.
- You are under pressure to move exams back into the hall and want the cost, capacity and accessibility trade-offs first.
- You need your assessment measures to meet GDPR and EU AI Act requirements on evidence, explainability and human oversight.
- You want a practical way to match verification to the stakes of each assessment instead of securing everything at once.
Abstract
Almost four years after generative AI reached students, higher education still lacks a stable answer to how a grade can reflect what a student knows. This white paper reviews the measures institutions have reached for since 2022 – AI detection, process tracking, watermarking, the return to the exam hall, oral exams and online supervision – and finds that each has a half-life. Measures that judge the finished work produce signals rather than evidence, while those that control the room cannot secure a student wearing AI glasses or earbuds.
Building on four preconditions – evidence, fairness, privacy and accessibility – and what to expect by 2030, the paper argues for a shift from securing the product to verifying understanding: oral defences, staged and combined formats, AI-permitted tasks with a verification step and short checks of understanding at hand-in. It closes with a risk-based model for matching protection to the stakes of each assessment, and a three-stage path institutions can follow within the constraints they already work under.