AgID Guidelines on AI in Public Administration: What Changes for Public Bodies and Suppliers

The AgID guidelines on artificial intelligence in public administration were published on 9 November 2026 with Determination No. 169. The document tells public bodies how to classify AI systems, who governs them and how to manage their data, security and skills.¹² For IT suppliers, the result is a more precise national reference on how to design, present and manage solutions intended for the public sector.²

What the AgID guidelines provide

The Agency for Digital Italy (AgID) issued the 140-page text pursuant to Article 71 of the Digital Administration Code. The guidelines were provided for in the Three-Year Plan for IT in the Public Administration 2024-2026 and aim at a conscious, effective and responsible use of AI.¹³ The text was drawn up in the Consultation Table of the Three-Year Plan, with the participation of numerous bodies, including ANAC, ANCI, Consip, INPS, INAIL, ISTAT, MEF and the Conference of the Regions. It incorporates the AI Act, the European regulation on artificial intelligence, and national Law 132/2025.¹³ The first pillar is risk classification: every AI system must be placed within the levels set out by the AI Act. For high-risk solutions in particular, obligations of transparency, human oversight and clear attribution of responsibilities apply.¹³ The second is an adoption model based on the PDCA cycle (Plan-Do-Check-Act, i.e. plan, do, check, correct) and on the ISO/IEC 42001 standard, dedicated to AI management systems. It is accompanied by tools for self-assessing organisational and technological maturity.¹² The third concerns data and security. Data quality, understood as accuracy, consistency, completeness and timeliness, is a prerequisite for reliable systems. Cybersecurity measures must counter threats specific to AI: evasion, poisoning, privacy and abuse attacks.¹³ The fourth is training. The guidelines promote a person-centred model of AI, with a code of ethical conduct, investment in digital literacy and the development of specialised profiles in the Digital Transition Offices (UTD).¹³

Strategy and governance in public bodies

Each administration must adopt an AI strategy in line with its own mission, with objectives, actions and ways to ensure transparency. This strategy must be aligned with the Integrated Plan for Activities and Organisation (PIAO), with data management rules and with the Three-Year Plan: AI thus enters the planning already in place, without a separate document.³ Coordination of adoption is entrusted to the Head of Digital Transition (RTD) and the relevant UTD. Before starting, public bodies must analyse their technological and organisational capabilities, data availability, staff skills and financial resources.³ AgID urges bodies to favour open source or open weight models, with a centralised orchestrator architecture that integrates different models according to need. For smaller administrations it encourages experimentation and shared projects, to prevent AI from widening the gap with better-equipped bodies.³

What changes for IT suppliers

According to AI4Business’s reading, the guidelines put the life cycle, verifiability and operating costs at the centre, while the decision remains in the hands of a person, as Law 132/2025 provides. Those selling to the public administration will therefore have to show how the product integrates with the body’s systems, how it is controlled, how it handles data and how it is monitored, in addition to the model’s performance.² Consider a municipality introducing an assistant for citizens: it must check the quality of the answers, and it must also assess the data used, the point at which a human operator steps in and who takes responsibility for errors.² The text refers to the AI Act timeline. The rules on literacy and the prohibitions apply from 2 February 2025, and many transparency and governance provisions from 2 August 2026. For the high-risk systems of Annex III the general deadline has been moved to 2 December 2027, and for those of Annex I to 2 August 2028.²

Why it matters

The document aims to move administrations from experimentation to systematic use, treating AI as a system to be governed and not as a software purchase.³² For suppliers, the requirement to document tests, changes and responsibilities, according to AI4Business’s reading, could shift the comparison from the product’s performance to the ability to manage it over time. How far this will translate into tender requirements depends on the calls for tender, because the guidelines are not equivalent to a tender specification.²

Do the AgID guidelines bring forward the AI Act deadlines?

No. According to AI4Business, the publication of the guidelines does not by itself bring forward all the AI Act deadlines: the text refers to the European timeline, which remains the reference.²

This article was written by artificial intelligence from the sources cited, without human review. Automatic newsroom by Redazia.

Sources

Footnotes

1. AgID pubblica le Linee Guida per l’adozione di IA nella Pubblica Amministrazione, Homepage | Agid, 2026-11-09 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

2. AI nella PA, le linee guida AgID cambiano i progetti pubblici, AI4Business, 2026-11-10 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

3. Intelligenza artificiale nella PA, ecco le linee guida AgID: come adottarla, gestirla e controllarla, CorCom, 2026-11-10 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹