On 11 March 2026, AgID published Version 1.0 of the draft guidelines, titled "Linee Guida per lo sviluppo di sistemi di Intelligenza Artificiale nella pubblica amministrazione", opened for public consultation. AgID states the document does not carry a financial penalty and is not a final binding instrument at this stage.
AgID sets out seven general principles drawn from Law 132/2025 that the guidelines state are applicable to the development and procurement of AI systems across the public administration. According to AgID, these principles cover: the protection of fundamental rights, transparency, proportionality and non-discrimination (Principle 1); the quality, accuracy, reliability and transparency of data and processes (Principle 2); human-centredness, explainability and harm prevention (Principle 3); the protection of democratic and institutional processes and sovereignty (Principle 4); coherence with the European regulatory framework, specifically the EU AI Act (Principle 5); cybersecurity and resilience across the system lifecycle (Principle 6); and universal accessibility and inclusion (Principle 7).
The draft guidelines also establish a reference logical architecture for AI systems in the public administration. AgID defines a layered AI stack encompassing an Energy Layer, a Chip Layer, an Infrastructure Layer, an AI Model Layer and an Application Layer. AgID further identifies a reference AI architecture model comprising AI models, heterogeneous data sources across cloud, on-premises and hybrid environments, application-level tools and an orchestrator component. AgID classifies AI system families to which the guidelines apply, including statistical or data-driven AI, machine learning, neural networks, deep learning and generative AI. AgID addresses the role of data in AI system development, including training data and operational data, as a distinct dimension of the reference architecture.