In 2026, the AI Act enters its truly “operational” phase for many organizations: it’s no longer just a conference topic, but a set of obligations and educational choices that directly affect schools and universities. For teachers, this mainly means three things: rethinking assessment (even when it’s online), clarifying the use of AI outside the classroom (off campus ai), and building a framework ofacademic integrity aithat is workable, fair, and defensible in case of disputes. This article takes a professional, teaching-focused angle: what changes, which risks to avoid, and which alternatives work in class and in exams.
AI Act 2026: what really changes for Italian schools and universities
When we talk aboutAI Act schooland its application in 2026, the point isn’t “whether” AI will be used, but “how” it will be governed. The AI Act introduces a risk- and responsibility-based logic across the value chain (provider, deployer/organizational user, etc.). In educational contexts, this translates into concrete choices: which tools to adopt, with which configurations, with which notices and controls, and with which internal procedures.
For schools and universities, the biggest impact shows up in three areas:
- Teaching and access: educational artificial intelligence tools used for tutoring, feedback, material generation, and study support require rules of use, attention to data, and transparency toward students and families.
- Assessment: proctoring, behavior detection, automated analysis of assignments, or systems that “classify” students can increase risk (privacy, bias, errors) and require more robust governance.
- Organization and accountability: AI regulation for universities and schools pushes institutions to clarify roles (who decides, who configures, who monitors), to document choices, and to train teachers and staff.
From a pedagogical standpoint, the positive side effect is that we start talking about design again: objectives, evidence of learning, assessment criteria, and coherence between in-class and at-home activities. In other words, the AI Act can become an opportunity to make explicit what often remains implicit: what we consider “learning” and what evidence demonstrates it.
Digital proctoring and assessment: risks, obligations, and compliant alternatives
Digital proctoring is often brought up as the “solution” to cheating in online exams. But it is also one of the most delicate cases: it combines surveillance, processing of personal data (sometimes biometric or behavioral), and automated or semi-automated decisions (flags, suspicions, risk scores). In theproctoring 2026perspective, this means: it’s not enough to “turn on a platform”—you need to show that the choice is proportionate, transparent, and controllable.
The main risks to consider (even before compliance) are educational and organizational:
- False positives and bias: movements, unstable connections, different home conditions (noise, shared spaces) can generate unfair flags, with consequences for trust and classroom climate.
- Transparency and disputes: if the student doesn’t understand why they were “flagged,” or if the teacher lacks tools to explain and verify, litigation increases and the legitimacy of the assessment decreases.
- Intrusiveness and accessibility: not everyone has adequate webcams, suitable environments, or the ability to be recorded; moreover, surveillance anxiety can worsen performance without improving learning.
If the school or university decides to use proctoring anyway, the guiding question should be:what is the evidence of validity and necessity with respect to the assessment objective?In practice, it’s worth setting up at least: activation criteria (when yes/when no), a clear notice, support channels, procedures for human review of flags, and alternative arrangements for those who cannot take the test under those conditions.
Many times, however, there are less invasive alternatives that are more consistent with authentic assessment (and often more effective against improper AI use):
- Short, frequent oral tests (even online) with variable questions, asking students to explain steps and choices: they shift the focus from the product to the reasoning.
- “Open resource” assignments designed to include sources and tools (including AI) but with assessment criteria focused on analysis, contextualization, citations, and limitations: they reduce the incentive to cheat because use is declared and guided.
- Process assessment: staged submissions (outline, draft, revision, final reflection) with micro-evidence. It’s a simple strategy that makes total substitution of the work harder.
In short: proctoring may look like a “technical” answer, but often the problem is one of design. In 2026, the most robust choice is to combineassessment design, transparency, and targeted oral checks, reserving digital surveillance for truly justified and well-documented cases.
Off Campus AI: how to manage AI use outside the classroom without losing academic integrity
Byoff campus aiwe mean the use of generative or assistive AI tools while studying at home: summaries, explanations, translations, brainstorming, style correction, example generation, up to the production of assignments. It’s the hardest area to “control” and, precisely for that reason, the one that benefits most from clear rules and well-designed activities.
A realistic approach starts from a distinction that is useful for students and defensible for teachers:legitimate supportvsimproper substitution. The first improves understanding and quality (e.g., asking for examples, having a concept explained, receiving feedback on a text). The second delegates to AI the cognitive decisions the task is meant to assess (thesis, argumentation, source selection, complete solution).
To reduce ambiguity and conflict, a “minimum policy” in three levels works well, communicated before the assignment:
- or
- and, if you want to try it with your class or course,
- . Responsible adoption is not a compliance exercise: it’s a teaching choice that, if done well, improves assessment quality and student autonomy.
On the teaching side, the most solid strategy is to design assignments that “hold up” even if the student uses AI. Some examples applicable in high school and university:
- Context-anchored tasks: connect theory to a case discussed in class, a lab, a dataset provided by the teacher, or a specific text with targeted questions. AI helps, but it doesn’t replace the in-class experience.
- A request for meta-reflection: “what choices did I make and why,” “which alternatives did I discard,” “what limits do the sources have.” It’s a strong indicator of learning, hard to generate credibly without understanding.
- Versioned submissions: draft 1 (with notes), guided peer review, draft 2 with a changelog. Even digitally, this builds traceability without turning into surveillance.
In this way, AI use becomes an educational content area (critical literacy) and not just a disciplinary problem. And above all, it reduces dependence on often-unreliable “detection” tools: better to design for integrity than to chase violations.
Academic integrity in the age of AI: practical guidelines for teachers (high school and university)
An operational framework ofacademic integrity aiis not built with a generic ban (“AI is forbidden”), because it’s hard to verify and often inconsistent with reality: students will use it anyway, at least to study. What works instead is a teaching pact based on clarity, process traceability, and coherence between objectives and evidence.
Here is a set of “ready-to-use” practices you can adapt for departments, class councils, and university courses (also useful for aligning with university AI regulation and institutional policies):
- Standard disclosure: add to every assignment a “AI Use” section with three fields: tool used, main prompts or instructions, parts of the work influenced by AI. Assess the quality of the disclosure (not just its presence).
- Lightweight process traceability: require 1–2 “pieces of evidence” (outline, concept map, revision log, reading notes). It’s not surveillance: it’s documentation of learning.
- Thinking-oriented rubrics: give more weight to criteria such as conceptual accuracy, quality of argumentation, critical use of sources, internal coherence, ability to connect with lessons/experiences. “Well-written text” alone is no longer enough as an indicator.
- Confirmation interviews (short viva): on samples of assignments or in doubtful cases, 3–5 minutes to ask the student to explain a choice, a source, a passage. It’s a proportionate measure that reduces false positives.
On “AI detectors”: as teachers, it’s important to know that there are no infallible tools to determine whether a text was generated by a model. Use them, if at all, as weak signals to start a conversation, not as proof. Proper handling is centered on observable criteria (process, consistency with previous performance, ability to defend choices) and on fair procedures.
How StudierAI can help: policy, training, and tools for responsible adoption
For many institutions, the critical point isn’t “finding an AI tool,” but building a sustainable ecosystem: clear policies, coherent assignments, teacher training, and tools that help students and teachers work transparently.StudierAIwas created in this direction: to support the responsible adoption of AI in educational contexts, with a specific focus on integrity, study method, and assessment practices.
In practice, effective support for teachers and institutional governance can include:
- Policy models and assignment “clauses”: ready-to-adapt texts to clarify permitted/not permitted use, disclosure, and assessment criteria. It reduces ambiguity and makes handling problematic cases more defensible.
- Training pathways for teachers: how to design “AI-resilient” assessments, how to evaluate process and product, how to set up rubrics and confirmation interviews, how to teach critical use of sources and the limits of models.
- Assignment and rubric templates: structures that ask for process evidence (outline, revisions, reflection) and make AI use an explicit, assessable element, not a “secret” to uncover.
If you are setting up or updating an internal policy on AI, proctoring, and at-home activities, the goal is to arrive at a repeatable practice: assignments with disclosure, coherent rubrics, and proportionate verification methods. You can explorewho we areorstart for freeand, if you want to try it with your class or course,sign up for free. Responsible adoption is not a compliance exercise: it’s a teaching choice that, if done well, improves assessment quality and student autonomy.
