With Determination no. 169 of 9 November 2026, the Agency for Digital Italy (AgID) adopted the final Guidelines for the adoption of artificial intelligence in the Public Administration, issued according to the procedure provided for in Article 71 of the Digital Administration Code and as part of the Three-Year Plan for Information Technology in the PA 2024-2026.
The document introduces an organic framework of organizational, technical and regulatory compliance requirements that administrations must consider when introducing and using AI systems, even when acquired from external suppliers.
The Guidelines first of all require administrations to define a strategy for AI, consistent with their institutional functions, with the Integrated Plan of Activities and Organization (PIAO) and with the data governance strategy. The strategy must identify objectives, areas of application, resources and transparency measures. Each project must also be accompanied by the identification of requirements, risk assessment and the definition of indicators to measure the effectiveness, reliability and organizational impact of the systems. The approach follows a model of continuous improvement, inspired by the Plan-Do-Check-Act cycle and the ISO/IEC 42001 standard, to be applied in a manner proportionate to the characteristics and complexity of the solutions adopted.
In terms of responsibilities, the coordination of AI governance is entrusted to the Digital Transition Officer (RTD) and the related Office, with the involvement of the Data Protection Officer (DPO) and cybersecurity contact persons. Administrations characterized by a particularly intense use of AI can also identify a dedicated Artificial Intelligence Manager. The adoption of procedures to govern the development, use and monitoring of systems is required, while the preparation of a specific code of ethics and conduct is an optional, albeit recommended, measure.
Of particular importance are the obligations on the protection of personal data, IT security and the protection of fundamental rights. Administrations must verify the lawfulness of processing, the quality and origin of data, the methods of use of personal information and any transfers to third countries. They must also carry out the impact assessments required by the GDPR and, in the cases provided for by the AI Act, those on fundamental rights, ensuring adequate human oversight mechanisms. Security must also include AI-specific risks, such as manipulation of training data, attacks on models and unauthorized extraction of information.
The operational implications also directly concern companies that develop or provide AI systems to public administrations. Entities will have to be able to acquire and verify the technical documentation, safety features, operating methods and evidence necessary to demonstrate the compliance of the solutions, including through checks after their introduction. It is therefore essential for administrations and suppliers to clearly define responsibilities, security requirements, surveillance methods, information obligations and verification tools throughout the entire life cycle of the systems.
The Guidelines expressly distinguish between mandatory requirements, prohibitions and recommendations. Their adoption therefore represents an important step towards a structured, documentable and continuously verified governance of AI in the PA, destined to affect both the internal organization and the methods of acquisition and management of technologies.