DM 219 PNRR and AI in schools: what teachers must decide by summer 2026

Between 2025 and 2026, the adoption of artificial intelligence in schools stops being an individual choice “of the single teacher” and becomes a system-wide issue: rules, assessment, privacy, training, tools. The operational reference many schools are using to find their bearings isMinisterial Decree 219 2025 PNRR school, with PNRR deadlines pushing decisions to be made by summer 2026 in order to be ready for the 2026–2027 school year.

This article translates the regulatory and project framework into concrete choices: what to approve in the teachers’ board, what to define in subject departments, what to put into practice in class councils. The goal is not to “make an AI regulation,” but to build a sustainable teaching framework: meaningful assignments, robust assessments, shared integrity criteria, and training that truly improves teaching.

DM 219/2025 and PNRR: what really changes for teachers (and why the deadline is summer 2026)

In school practice, “what changes” does not coincide with a single circular, but with a set of constraints and opportunities: funding, reporting, innovation targets, and above all the need to make PTOF, curricula, assessment, and digital tools consistent with one another. In this scenario, DM 219/2025 is often read as an accelerator: it pushes schools to structure actions on digital skills and the conscious use of technologies, including AI as a cross-cutting theme (teaching, digital citizenship, inclusion, assessment).

Why is summer 2026 a turning point? Because many schools are planning within that window: (1) the closing of PNRR-related design/training cycles, (2) the update of the PTOF and internal regulations, (3) the definition of assessment procedures and the handling of cases of improper AI use. Reaching September 2026 without shared decisions means leaving each class “on its own” to manage tools that students already use every day.

By spring–summer 2026, the operational decisions that typically fall to the different bodies are:

• Teachers’ board: common guidelines on AI use, transparency criteria, an assessment framework, training priorities, and integration into the PTOF.
• Departments: subject-specific standards (what is allowed in assignments, how to cite AI, which tasks remain “by hand,” which become hybrid), rubrics, and authentic tasks.
• Class councils: adaptation for each class group, communication to families, management of PDP/PEI and equity measures (e.g., compensatory tools vs undue advantages).

The teaching point is clear: AI is already “in the context.” The choice is not whether to let it in or not, buthow to govern itto maintain the validity of assessments, the quality of learning, and trust in grading.

School-wide rules on AI use: what to define now to avoid chaos in 2026–2027

An effective policy is not a list of bans: it is an educational pact that makes expectations and responsibilities explicit. If the school does not define a minimum framework, every teacher will be forced to invent rules “in an emergency,” with side effects: inconsistency across subjects, conflicts with students and families, disputes over grades and sanctions.

For the 2026–2027 year, the best-functioning schools are converging on four pillars, all of which can be translated into regulations and teaching instructions:

• Areas of use: what is allowed (e.g., brainstorming, clarifications, guided exercises, language revision) and what is forbidden (e.g., fully generating graded work without disclosure, solving ongoing tests).
• Transparency: the obligation to declare whether and how AI was used (prompt, steps, output, edits), with simple, standard wording.
• Traceability and privacy: approved tools, accounts, data handling, and a ban on uploading sensitive content (personal data, PEI/PDP, non-public tests) to unauthorized platforms.
• Assessment consistency: guidance on homework, in-class tests, oral exams, and on how AI affects (or does not affect) the grade.

A specific issue, often overlooked, concernsAI at school 2026 2027 rulesfor homework: banning it outright is unrealistic; allowing it without conditions makes assessment fragile. The most solid solution is to distinguish between “practice” tasks (where AI can be a tutor, with disclosure) and “assessment” tasks (where constraints are needed: in-class production, oral, process portfolio, or submission with work traces).

Include in the policy also a one-page “operational” appendix with examples: how to cite AI, how to attach prompts and revisions, how to handle AI errors (hallucinations) as an opportunity to check sources. This reduces arbitrariness and makes the rule teachable, not only punishable.

Plagiarism, off-campus AI, and academic integrity: practical assessment criteria and “AI-resilient” tests

With generative AI, the problem is not only “copying.” It is redefining the line between legitimate support and the replacement of cognitive work. In particular, teachers ask for criteria onstudent AI use off campus ai plagiarism: what happens when the work is produced at home with an AI assistant and arrives in class “perfect” but opaque?

Three practical guidelines help manage integrity without turning school into a courtroom.

1) Shift the focus from the “product” to the “process.” When the process is assessed (outlines, drafts, argumentative choices, sources, self-corrections), AI becomes a visible tool. This is consistent with pedagogical evidence on formative assessment and metacognition: learning improves when the student makes strategies and revisions explicit.

2) Define clear thresholds of “permissible help.” A useful example (adaptable by age and track):

• Permissible: asking for examples, explanations, self-check questions, improving linguistic clarity while keeping one’s own ideas, generating a concept map to compare with the textbook.
• To be disclosed and discussed: reorganizing a text, proposing an outline, suggesting counterarguments, simulating an oral exam with feedback.
• Impermissible (if not authorized): generating the complete assignment, solving graded exercises without understanding, producing “ready-made” answers for tests or oral exams.

3) Design “AI-resilient” tests. This does not mean tests that are impossible to copy, but tests where copying is not worth it because situated understanding is required. Some effective formats in secondary school:

• If you want to explore the tool in an operational way, you can
• and evaluate within the department which assignments and which rubrics make the use didactically transparent. To learn more about the approach and the project context, you can also find the page
• .

Bringing AI into school seriously, by summer 2026, means closing three workstreams: a clear policy, AI-resilient assessment, and training with concrete outputs. This way AI becomes an ally for better learning, not a source of conflict or distrust. 2026–2027 will reward schools that chose method and coherence, not those that chased the latest tool.academic integrity ai cheating high school: not a “witch hunt,” but education in responsibility (use disclosure), proportionality (graduated sanctions), and repair (retaking the test, reflective interview, process-based submission).

Finally, be careful with detectors: they are fallible and can penalize students with “clean” writing styles or non-native Italian. Use them, if at all, only as a weak signal. Handling doubtful cases should be based on an interview, a request to explain the steps, comparison with previous work, and in-person verification. It is a cultural shift: from the “surprise” test to the “argued and defensible” test.

Mandatory training and governance: how to design useful pathways (not just compliance) with PNRR funds

Many schools are planningartificial intelligence training for teachers 2026with PNRR funds. The risk is to focus everything on “tool tutorials” (which change in a few months) and too little on teaching, assessment, and design. A useful pathway, instead, starts from transferable skills and produces evidence: learning units, rubrics, authentic tasks, examples of responsible prompts, and transparency grids.

A minimal governance setup (realistic for a school) can include:

• An AI working group (digital animator + department leads + assessment function lead) with a clear mandate: policy, training, monitoring.
• Measurable objectives: e.g., by June 2026 each department produces 2 AI-resilient tests and 1 rubric with transparency criteria; by September 2026 each class council agrees on a standard use-disclosure requirement.
• Micro-training in practice: workshops on real tasks (tests, oral exams, homework), with peer review and classroom experimentation.
• Light monitoring: collection of examples, critical issues, integrity cases, and annual policy updates (not every week).

To prevent training from remaining abstract, tie it to three guiding questions: (1) which activity do I improve tomorrow in class? (2) which risk do I reduce (privacy, cheating, bias)? (3) which evidence do I bring (rubric, test, unit, protocol)? This approach makes training a teaching investment, not just a compliance task.

In parallel, update the PTOF and curricula with shared language: digital citizenship skills, critical use of sources, error management, and the ability to document processes. These are skills that remain valid even when tools change.

How StudierAI can support teachers and students in a compliant way: use cases and criteria for choosing the platform

Once rules and criteria are defined, you need a tool that makes them workable.StudierAIcan be included in a compliant framework if it is used as support for studying and metacognition, not as a shortcut to “produce assignments.” The difference is made by the instructions: what the student is asked to do with the output and what they must disclose.

Examples of didactically solid use cases (and easy to make transparent):

• Guided summaries: the student compares the summary with the text and flags 3 missing or inaccurate points, citing the page of the book or the class notes.
• Flashcards and quizzes: use for retrieval and consolidation, with an “error log” submission (which questions were wrong and why).
• Oral simulations: the student practices, then brings to class a map with 5 “difficult” questions that emerged and the reasoned answers (not copied).
• Study planner: realistic workload planning, especially useful in classes with organizational difficulties; the teacher can ask for a brief reflection on what worked and what didn’t.

For teachers, the advantage is turning AI into an “amplifier” of well-known practices: spaced retrieval, the testing effect, frequent feedback, active study. These are evidence-based principles: they do not depend on technological fashion and integrate well with formative assessment.

When does a tool become risky instead? When it encourages opacity (final outputs without process), when it does not clarify responsibilities, or when it pushes students to upload sensitive materials. Here comes the key question for every school:how to choose compliant AI platforms for teaching?

An essential checklist (to use in the digital committee or AI working group) includes:

• Privacy compliance and data management: where data flows, which logs are kept, the possibility of minimization and control.
• Use transparency: the ability to document prompts, steps, revisions, or at least to guide the student to declare use in a standard way.
• Alignment with the policy: does the tool support studying, self-checking, and feedback? Or does it encourage “turnkey” production of assignments?
• Accessibility and equity: it works on different devices, does not create economic barriers, and lends itself to inclusive uses (timing, modes, supports).

If you want to explore the tool in an operational way, you canstart for freeand evaluate within the department which assignments and which rubrics make the use didactically transparent. To learn more about the approach and the project context, you can also find the pagewho we are.

Bringing AI into school seriously, by summer 2026, means closing three workstreams: a clear policy, AI-resilient assessment, and training with concrete outputs. This way AI becomes an ally for better learning, not a source of conflict or distrust. 2026–2027 will reward schools that chose method and coherence, not those that chased the latest tool.