AI Act and schools: what changes for proctoring and Off-Campus AI

The entry into force of the AI Act marks a concrete turning point for schools: no longer just “best practices” on the use of artificial intelligence, but requirements, responsibilities, and controls that directly affect teaching, assessment, and digital tools. Two areas are particularly sensitive for teachers: theA new (and often underestimated) point is theOral simulations: practice for presenting with progressively challenging questions and a request for examples; also useful for reducing anxiety and improving communication skills.

Planner and study plan: time management with realistic goals; didactically effective if the student revisits the plan and justifies the choices (metacognition).

To remain consistent with the AI Act and with expectations of transparency, the school can adopt a simple model: (1) define activities where AI is allowed and with what limits; (2) require a usage declaration; (3) assess primarily reasoning and application; (4) provide a brief verification interview when needed. This way AI becomes a support for studying, not a substitute for performance.Data: what should never be uploaded (third parties’ personal data, documents with names, certifications, health information, credentials)., the practical point is full applicability: many provisions become effective gradually, but the threshold that most concerns schools and universities isIf you are building school-wide guidelines, it can also be useful to share the rationale behind the choices: not “we allow/ban AI,” but we teach how to use it responsibly, protecting data and ensuring fairness in assessment. It’s a message that reduces conflict and increases adherence to the rules, because it makes the educational purpose visible., when the framework of obligations for various high-risk systems and for actors in the supply chain is fully operational. For teachers this doesn’t mean “becoming lawyers,” but knowing which questions to ask and which safeguards to demand when an institution adopts AI-based tools.

who we are. In a context where rules evolve, the most effective choice for teachers is to keep the method steady: transparency, proportionality, evidence of learning, and the centrality of the educational relationship.(who develops/provides the system),Instructional tracking: a requirement to retain process evidence when the assignment calls for it (drafts, main prompts, revisions, sources consulted).(who uses it in their own context: school, university, training provider), and sometimes intermediaries or integrators. An institution that adopts a proctoring system or that mandates/recommends a study platform with AI agents becomes part of the chain of responsibility: it must verify that the service is compliant, that use is proportionate, and that clear information exists for students and families.

Why are proctoring and Off Campus AI among the most sensitive cases? Because they combine three factors:Academic integrity: between AI detection, assessment design, and a culture of responsible use(with consequences for grades and pathways),The topicacademic integrity ai detectionis often approached with a conditioned reflex: “we’ll use a detector and solve it.” In reality, detectors have known limits: false positives (simple styles, non-native speakers, heavily reworked texts) and false negatives (well-written human texts or AI that has been post-edited). For a teacher, the biggest risk is turning a probabilistic indicator into certain proof, with disciplinary consequences that are hard to sustain.(students who don’t really know how the algorithm works or how logs are used). Moreover, schools work with vulnerable populations (minors) and in contexts where trust is an educational prerequisite: an error or a perception of unfairness can undermine classroom climate and motivation.

assessment design

process assessmentanda culture of responsibility. In practice, this means making the learning pathway visible and rewarding skills that AI cannot “replace” without leaving traces: situated reasoning, argued decisions, connections to specific lessons and sources.,Tools and routines that work well in the classroom (and that don’t require invasive technologies):andExplicit rubrics: criteria on sources, coherence, originality of viewpoint, personal/lab examples, and quality of revision.: neurodivergent students, students with disabilities, with different home environmental conditions, or with unstable connectivity can be penalized by “anomalous” signals that do not indicate cheating.

From an organizational standpoint, if an institution is considering adopting proctoring, at least four levels of instructional-managerial control are needed (even before the technical ones):

• Proportionality: use proctoring only when the stakes justify it (certifying exams, tests with legal value), avoiding normalizing it for routine assessments.
• Transparency toward students: explain what is monitored, for what purposes, how long data are retained, and how to contest decisions or flags.
• Human oversight: no automatic sanction or invalidation; flags must be clues, not verdicts, and must be read in context (accessibility, environment, exam stress).
• artificial intelligence education 2026

On the instructional side, the most useful question is: can we reduce the need for surveillance by increasing the quality of the assessment? Evidence on assessment suggests that authentic tasks, clear criteria, and triangulation of evidence reduce both the opportunity and the incentive to cheat. Feasible alternatives (even on tight timelines) include:

• Short, spot-check oral follow-ups after a written test: 3–5 targeted minutes on choices, steps, and sources used.
• StudierAI
• start for free
• sign up for free

These solutions don’t eliminate the need for rules, but they shift the center of gravity from surveillance to design: an approach more consistent with the educational mission and more robust against disputes and false positives.

Off Campus AI and AI agents for studying: governance rules, data, and responsibilities

ByFlashcards and review: generating Q&A for spaced repetition, with teacher oversight on accuracy and cognitive level (recall vs application).we mean the use of AI tools outside institutional environments: at home, in the library, on a personal phone. Here the school often doesn’t “choose” the technology, but bears its effects. In 2026, with growing attention to transparency and accountability, it’s worth moving from a generic ban to clear governance: what is allowed, what must be declared, which data must not be entered, how risks are managed.

A new (and often underestimated) point is theOral simulations: practice for presenting with progressively challenging questions and a request for examples; also useful for reducing anxiety and improving communication skills.: not only chatbots that answer, but systems that plan, retrieve materials, propose quizzes, simulate oral exams, and sometimes act autonomously (e.g., organizing a study plan or generating successive versions of an assignment). This increases usefulness, but also risks: cognitive dependence, opacity of sources, and inadvertent data sharing (notes with sensitive information, documents with names, grades, diagnoses, etc.).

A “minimal but effective” policy for off-campus use should include five operational decisions, communicable on one page and referable in assignments:

• Data: what should never be uploaded (third parties’ personal data, documents with names, certifications, health information, credentials).
• If you are building school-wide guidelines, it can also be useful to share the rationale behind the choices: not “we allow/ban AI,” but we teach how to use it responsibly, protecting data and ensuring fairness in assessment. It’s a message that reduces conflict and increases adherence to the rules, because it makes the educational purpose visible.
• To explore the approach and the project’s philosophy in more depth, you can also consult
• . In a context where rules evolve, the most effective choice for teachers is to keep the method steady: transparency, proportionality, evidence of learning, and the centrality of the educational relationship.
• Instructional tracking: a requirement to retain process evidence when the assignment calls for it (drafts, main prompts, revisions, sources consulted).

This governance is not meant to “control” students’ private lives, but to make AI use compatible with educational goals: autonomy, ability to argue, and information literacy. In other words: AI can be a study accelerator, but only if the teacher defines what counts as observable learning and as an assessable product.

Academic integrity: between AI detection, assessment design, and a culture of responsible use

The topicacademic integrity ai detectionis often approached with a conditioned reflex: “we’ll use a detector and solve it.” In reality, detectors have known limits: false positives (simple styles, non-native speakers, heavily reworked texts) and false negatives (well-written human texts or AI that has been post-edited). For a teacher, the biggest risk is turning a probabilistic indicator into certain proof, with disciplinary consequences that are hard to sustain.

A more instructionally sound approach combines:assessment design,process assessmentanda culture of responsibility. In practice, this means making the learning pathway visible and rewarding skills that AI cannot “replace” without leaving traces: situated reasoning, argued decisions, connections to specific lessons and sources.

Tools and routines that work well in the classroom (and that don’t require invasive technologies):

• Explicit rubrics: criteria on sources, coherence, originality of viewpoint, personal/lab examples, and quality of revision.
• “AI-proof” assignments: requiring integration of materials provided by the teacher, citing specific passages, or responding to constraints (method, steps, justification of choices).
• AI usage declaration: a standard final section (2–5 lines) on how it was used and what was verified by the author.
• Versioning: submission of a draft + revision with a metacognitive comment (what I changed and why).
• Targeted oral checks: questions on a critical step, on a source, or on an applied example; excellent also as micro-interviews.

Detectors can remain as a weak signal (to decide whether to do a clarification interview), but the main lever is design. This approach is also consistent with the horizon ofartificial intelligence education 2026: not a “witch hunt,” but literacy in the critical and responsible use of tools that students will use anyway.

How StudierAI can support teachers and students in a compliant way (summaries, flashcards, quizzes, oral simulations)

In practice, many teachers look for tools that improve studying without turning into opaque “shortcuts.” In this senseStudierAIcan be included in a clear learning agreement: AI as support for understanding, retrieval, practice, and oral preparation, with rules on transparency and data protection. If you want to explore it with a pilot class, you canstart for freeorsign up for freeand define from the outset how to document use in assignments.

Here are some didactically “clean” use cases, i.e., oriented toward learning and compatible with a culture of integrity:

• Guided summaries: the student turns notes or materials into a synthesis with constraints (length, keywords, check questions). Useful if paired with a request to “verify sources” and to provide personal examples.
• Flashcards and review: generating Q&A for spaced repetition, with teacher oversight on accuracy and cognitive level (recall vs application).
• Quizzes and self-assessment: sets of exercises with feedback; excellent for remediation and consolidation, provided the student notes mistakes and the reasoned correction.
• Oral simulations: practice for presenting with progressively challenging questions and a request for examples; also useful for reducing anxiety and improving communication skills.
• Planner and study plan: time management with realistic goals; didactically effective if the student revisits the plan and justifies the choices (metacognition).

To remain consistent with the AI Act and with expectations of transparency, the school can adopt a simple model: (1) define activities where AI is allowed and with what limits; (2) require a usage declaration; (3) assess primarily reasoning and application; (4) provide a brief verification interview when needed. This way AI becomes a support for studying, not a substitute for performance.

If you are building school-wide guidelines, it can also be useful to share the rationale behind the choices: not “we allow/ban AI,” but we teach how to use it responsibly, protecting data and ensuring fairness in assessment. It’s a message that reduces conflict and increases adherence to the rules, because it makes the educational purpose visible.

To explore the approach and the project’s philosophy in more depth, you can also consultwho we are. In a context where rules evolve, the most effective choice for teachers is to keep the method steady: transparency, proportionality, evidence of learning, and the centrality of the educational relationship.