From 2026, the debate on AI and assessment will no longer be only “methodological” or “technological”: it also becomes a matter of organizational responsibility. The AI Act pushes schools and universities to demonstrate that the use of artificial intelligence systems is transparent, proportionate, and protective—especially when it enters assessment and surveillance processes. The point is not to demonize AI, but to shift the center of gravity: less invasive control, more instructional design, more evidence, and more traceability of choices.
This article is intended for teachers and coordinators: it clarifies what the full entry into force of the obligations from 2 August 2026 entails operationally (AI Act school 2026), why proctoring is among the most critical cases (AI Act university proctoring; university exam proctoring 2026), and how to redesign assessment through the lens of academic integrity artificial intelligence, with strategies applicable in the classroom and in university courses.
AI Act 2026: what changes for schools and universities from 2 August
From 2 August 2026, obligations and responsibilities increase in a concrete way for anyone using AI systems in education, especially when AI affects decisions, assessments, or access to services. For teachers, the key message is simple: it is not enough that a tool “works”; you need to be able to explainhowit works,whyit is used andwhat safeguardshave been put in place for students and teachers.
In practice, the AI Act regulation for teachers translates into three operational areas that affect day-to-day teaching:
- Transparency towards students: when an AI system is used to support learning or assessment, it is necessary to clearly inform what it does and what it does not do (e.g., suggests feedback, generates quizzes, detects anomalies), and what data it processes.
- Governance and accountability: an internal decision chain is needed (who authorizes, who monitors, who is responsible). In a department this means: shared policies, roles (teaching lead, DPO/privacy office, IT support), and update procedures.
- Protection and contestability: if AI influences outcomes (directly or indirectly), it is necessary to guarantee the possibility of clarification, review, and challenge. Assessment cannot become a “black box” that the student is subjected to.
A useful way to orient yourself is to ask: is AI here only a study tool (low impact) or does it enter the assessment/disciplinary process (high impact)? The closer you get to decisions with consequences (eligibility, grade, suspicion of fraud), the more documentation, human oversight, and risk-minimization measures are needed. This change in perspective is central to AI Act school 2026, because many practices born “in an emergency” (distance learning, improvised online exams) now need to be rethought with stable and verifiable criteria.
Proctoring in exams: why it falls among the most critical cases (and what universities risk)
Proctoring (remote monitoring during exams) is often presented as a “technical” solution to the integrity problem. In reality it is a socio-technical system that combines data collection (video, audio, screen, metadata), automated analysis (reports of “anomalies”), and human decisions (validation, sanctions, retaking the test). Precisely for this reason, within the AI Act university proctoring framework, it is among the most delicate cases: it can impactprivacy,biasand the dynamics ofautomated decision-making(even when we are “only” talking about scoring or flags).
In 2026, with university exam proctoring 2026, it becomes harder to defend practices such as:
- Continuous and generalized surveillance as the default, without a clear analysis of necessity and proportionality with respect to educational objectives.
- Opaque automatisms: “the system detected that…”, without explainable criteria and without evidence that can be verified by the teacher and challenged by the student.
- Excessive data collection: unnecessary recordings, long retention, sharing with non-essential third parties, or secondary uses (training/analytics) that are not clarified.
What do universities risk? Beyond legal and reputational implications, the educational risk is often underestimated: when assessment is perceived as surveillance, anxiety increases, the relationship of trust worsens, and students are pushed to “beat the system” instead of learning. The literature on formative assessment and motivation shows that high-threat contexts reduce self-regulation and encourage superficial strategies. In other words: proctoring can protect the exam, but weaken learning.
Operationally, if a university decides to maintain forms of proctoring, it must prepare an evidence dossier: system description, data flows, flagging criteria, anti-bias measures, human oversight, complaint procedures, retention times, and the supplier’s responsibilities. This is not bureaucracy for its own sake: it is what enables the teacher to defend the fairness of the assessment when it is called into question.
How to redesign assessment without invasive controls: academic integrity strategies
The useful question is not “how do I prevent cheating?”, but “how do I design a test in which cheating has little value and you can see who can apply knowledge?”. This is the logic of academic integrity artificial intelligence: acknowledging that generative tools exist and shifting assessment toward performances that require context, reasoning, responsibility, and process traceability.
Concrete strategies (combinable) that reduce dependence on proctoring:
- Authentic and situated tasks: cases, scenarios, datasets, real documents from the discipline. Require justified decisions (not just answers). Example: “Choose among three strategies, justify trade-offs, indicate risks and selection criteria.”
- Short structured orals: 8–12 minutes with a rubric, follow-up questions, and a request for personal examples (how you solved it, why). Increases reliability and reduces arbitrariness.
- Well-designed open-book exams: allow resources (including AI) but assess application, comparison between sources, quality of assumptions. Add constraints: cite what was used, justify choices, include a “limits and checks” section.
- Analytic rubrics: criteria visible before the test (accuracy, reasoning, evidence, communication, critical reflection). Rubrics make it easier to defend assessment and reduce conflicts.
- Multiple versions and smart randomization: same objective, different data or parameters. Useful in numerical quizzes, programming, translation, analysis exercises.
- Multi-step application questions: ask not only for the result, but for the path (choices, checks, alternatives). AI can help with writing, but it is harder to simulate coherence and verification if the task is well designed.
Alongside test design, a clear policy on AI use is needed—short and workable. An effective model for courses (secondary school and university) includes three levels:
- Allowed: brainstorming, clarifications, generating examples, self-assessment, language improvement, provided the student remains responsible for the content.
- Conditional: use of AI in open-book tests or project work with a declaration requirement (what I used, for what, and how I verified it).
- Prohibited: generating entire assignments when the goal is to assess a specific individual competence (e.g., argumentative writing without aids) or to circumvent instructions and constraints.
This clarity reduces conflicts and, above all, makes it possible to teach skills that are essential today: citation, source verification, error checking, and authorial responsibility. These are assessable skills, and they turn AI from a threat into a teaching object.
“Safe” AI study tools instead of proctoring alone: the role of StudierAI
If the goal is to reduce the pressure on control, the most effective lever is to increase the quality of preparation. AI-powered study support platforms, if designed responsibly, help build learning routines and make assessment less dependent on invasive measures. In this senseStudierAIpositions itself as a teaching-oriented tool: summaries, flashcards, quizzes, oral simulations, and review activities that can be integrated into the course as a “gym” before the test.
For a teacher, the value is not “making students use AI,” butguidingthe use of AI toward activities consistent with learning outcomes. Here are four applicable examples (school and university):
- Diagnostic pre-test: before a module, assign a short quiz to surface misconceptions. In class you work on typical errors, not on what “seems clear.”
- Spaced review with flashcards: useful for content-heavy subjects and technical vocabulary. Reduces pre-exam “cramming” and improves long-term retention.
- Oral simulations: have the student practice explaining concepts and answering follow-up questions. In the exam setting, the structured oral becomes fairer because it trains performance, not just memory.
- Application exercises with feedback: multi-step questions and a request for justification. Immediate feedback supports self-regulation and makes it harder to “delegate everything” to AI.
The educational point is that these tools shift attention from “surveillance” to “verifiable preparation”: more guided practice, more clarity on objectives, more opportunities for feedback. If you want to explore a quick workflow, you canstart for freeorsign up for freeand evaluate how to integrate study activities consistent with your exam tasks.
Checklist for teachers and departments: what to do now to be ready for 2026
The transition doesn’t happen in July 2026: it is built now, with small but documented steps. Below is an essential checklist, designed for teachers, course coordinators, and departments. The goal is to arrive at an assessment system that holds up educationally and in terms of compliance, reducing dependence on proctoring.
- Map where AI enters teaching: tools used by students, tools adopted by the course, institutional tools. Distinguish study support vs assessment/surveillance.
- Update the exam policy: what is allowed, what is prohibited, what is conditional. Add practical examples and a section on declaring AI use in permitted assessments.
- Redesign at least one “high-integrity” assessment per course: authentic, well-constrained open-book, structured oral, or a project with milestones. Measure the effect (quality of submissions, grading time, complaints).
- If you use proctoring: request documentation from the supplier (data flows, flag criteria, bias mitigation, retention, subcontractors). Define what the teacher does and what the platform does, and put in writing that the final decision is human and reasoned.
- Notice and consent (when necessary): check texts, timelines, and channels with the privacy office/DPO. Avoid generic forms: clarity is needed on purposes, data, timelines, rights, and contacts.
- Audit data flows: where do recordings, logs, submissions end up? Who accesses them? For how long? Data minimization is an educational and organizational choice, not only a technical one.
- Challenge procedures: define how a student can request review of a flag or an outcome, what evidence is evaluated, and within what timelines. Reduces conflict and increases trust.
- Training for teachers and tutors: micro-modules on open-book design, rubrics, application questions, and responsible AI use. The goal is consistency across courses, not individual “heroics.”
The direction of travel is clear: from “control” to “design.” If assessment is well built, AI becomes content to teach (critical use, verification, citation) and digital tools become allies of learning. For those who want to explore the pedagogical approach and the principles of responsible development, it may also be useful to readwho we areand consider how to integrate AI tools for assessments without proctoring within a coherent course design framework.
