Note Notes 2026
The AI Act for Italian schools, universities and training centers
Translated from the Italian original.
In March 2026 the Italian Ministry of Education (MIM) launched the first national working group on AI in public schools, after the AI Act classified AI systems in education among the high-risk areas of Annex III. Institutions that run automated assessment, admissions, student monitoring, allocation of training resources fall within the strictest perimeter of Regulation (EU) 2024/1689. For universities, private schools, vocational training centers and edtech platforms, the practical consequence is imminent.
The point is that many centers have not yet realized they are inside the strictest regulatory perimeter.
What exactly counts as high-risk in education
To frame AI adoption in the Italian education system, the enterprise summary of the Stanford AI Index 2026 is useful. Annex III, point 3 of the AI Act lists four categories of educational AI systems classified as high-risk. Determining access or admission to educational institutions. Evaluating learning outcomes, including where the results are used to steer the learning path. Assessing the appropriate level of education for an individual. Monitoring and detecting prohibited behavior during tests.
Translated into concrete Italian cases. Software screening university course applications with AI scoring: high-risk. Proctoring systems for online exams that monitor the student via webcam with AI: high-risk. Adaptive learning platforms that decide a student’s level and path: high-risk. Automated assessment systems for written or oral assignments that carry weight in the student’s academic record: high-risk.
Teaching support tools that do not directly affect substantial decisions are outside the high-risk perimeter. An AI assistant that helps the teacher prepare materials is a productivity tool, not a decision-making system. An AI grammar checker for internal exercises is not high-risk if it does not determine a formal assessment.
The specific obligations for those using AI in education
Deployers of high-risk educational AI systems have heavy documentary and operational obligations. A risk management system throughout the life cycle. Guarantee of effective human oversight. Retention of automatic event logs. Transparency toward students and families about the use of the system. Periodic accuracy and robustness testing. Compliant technical documentation.
The analysis of how AI is changing work touches on related themes for teaching and administrative staff. For Italian universities the most sensitive point is human oversight. An AI system cannot on its own determine the admission or failure of a student. There must always be a human decision by someone who has seen the system’s output and evaluated it. This means that purely algorithmic admission procedures are not compatible with the AI Act without documented human review.
For vocational training centers the challenge is similar. If an AI system decides the entry level of a course or eligibility for a certificate, it must always be confirmed by a teacher or coordinator who has looked at the individual case. An automated procedure without effective human review does not meet the requirement.
AI proctoring: the most delicate area
AI proctoring is the category that will produce the most litigation. Proctoring platforms use AI to monitor students during online exams: face detection, gaze tracking, analysis of the home environment, identification of suspicious behavior. These systems process biometric data and have direct effects on the student’s academic record.
Even before the AI Act, the Italian Garante Privacy had issued critical decisions on proctoring. The 2022 GDPR ruling against a well-known online university placed strict constraints on the use of these systems: need for explicit consent, reasonable alternatives for those who do not want monitoring, minimization of the data collected.
With the AI Act the framework is strengthened. The system becomes high-risk, the university becomes a deployer with specific obligations, the student gains additional rights (clear information, the possibility to complain, effective human oversight of flags). Italian universities using these systems must update their policies and their contracts with suppliers by August 2, 2026.
Adaptive learning: a big opportunity, complex governance
Adaptive learning AI promises to personalize each student’s learning path based on their performance, pace and learning styles. Platforms such as Knewton, Smart Sparrow, the adaptive features of Cisco Networking Academy or Khan Academy are entering Italian training centers too.
When the system decides the student’s path, it is high-risk. The governance required is serious. Transparency on the personalization criteria (how the system decides what to offer the student). The possibility for the student or the teacher to override the suggested path. Periodic testing of effectiveness on different subgroups (to prevent personalization from producing learning bias). Retention of tracking data with clear legal bases and limited duration.
For Italian vocational training centers using adaptive learning, a minimum operational check requires: a supplier contract updated for the AI Act, an AI-specific privacy notice, a complaint path for the student, training of teaching staff on the use and limits of the system, an annual audit of effectiveness.
The opportunity: AI literacy training as an asset
Article 4 of the AI Act, applicable since February 2025, requires providers and deployers to ensure the AI literacy of staff working with AI systems. For schools and training centers this is a double opportunity: meeting the obligation for their own staff, and becoming recognized AI literacy providers for other organizations.
The European credential for AI professionals issued by AIPIA, the first EU-recognized AI training body in Italy to issue European Digital Credentials in this field, is becoming a reference for corporate and professional AI literacy programs. High schools and ITS institutes introducing AI courses into their curricula find in this framework a useful standardization reference.
Italian universities are making different choices. Some (Politecnico di Milano, Università Bocconi, Sapienza) have integrated AI literacy into mandatory courses and degree programs. Others are still at a preliminary stage. The difference will show over the next five years in graduate employability and in the attractiveness of universities on international markets.
The future of assessment: AI as support, not as decision-maker
AI assessment of students is a field that will evolve rapidly in the coming years. The direction emerging in the most advanced education systems (Singapore, Finland, some US states) is clear: AI as support for the teacher, not as a substitute for professional judgment. The teacher sees the AI output, evaluates case by case, decides.
This configuration is consistent with the AI Act. It is also pedagogically sounder. Assessing a written assignment is not just grammatical correctness or the presence of key concepts. It is professional judgment on the level of understanding, on the ability to argue, on the student’s individual development over time. All things a good teacher integrates differently from an algorithm.
The Italian edtech platforms positioning themselves intelligently sell AI as an efficiency tool for the teacher, not as a replacement. They cut the time spent correcting the mechanical parts and leave the professional time for the qualitative part. It is a model that works from a regulatory, pedagogical and commercial standpoint.
The deadline that really matters
August 2, 2026 for high-risk systems not governed by specific sectoral legislation. Educational systems fall under this date, with limited exceptions. This means that schools, universities and training centers using high-risk AI systems must be ready with documentation, governance, staff training and human review processes by that date.
Those starting today have 4-5 months. It sounds like a lot. For institutions with a hundred different systems in use, legacy contracts to review and staff to train, it is little. The operational choice that makes the difference is a complete mapping of the AI systems in use, to be done now, before operational pressure grows. Without mapping, any compliance plan is blind.
Training and education are sectors where AI can do remarkable things for students and teachers. But they are also sectors where people’s rights are delicate and the regulatory framework is rightly attentive. The Italian institutions that find the right balance between innovation and protection will be the ones that attract students, talent and investment in the next decade.