What the guidelines provide
On 9 November, AgID adopted the guidelines for the adoption of artificial intelligence in public administration. According to Agenda Digitale, the text is more prescriptive than the 2025 draft and assigns the digital transition manager two central tasks: AI management and human oversight of systems classified as high-risk.
The novelty therefore concerns the identification of a precise organizational control point within public bodies. This is not merely principle-based guidance on the use of algorithms, but the assignment of operational and supervisory functions to a figure already provided for in the public administration framework.
The break with Law 132/2025
The critical point flagged by the article is its relationship with Law 132/2025. According to the source, that law assigns responsibility for measures to the person who decides, that is, to the one who concretely adopts the act. On this specific aspect, the AgID guidelines do not reference the law: the alignment between the two figures therefore remains to be defined in each administration.
The question is relevant because the two roles may not coincide. The digital transition manager may be called upon to manage AI systems and exercise human oversight, while responsibility for the measure falls on the person who signs the decision. If the two people are different, it must be established who is accountable for what and through which documented steps.
What remains to be clarified
The available material does not indicate whether the guidelines contain specific provisions to harmonize the two profiles, nor whether common templates or decision-logging obligations are envisaged. The article merely observes that it is up to the entities to define procedures and alignments between roles that may not coincide.
A second element to verify concerns the scope of application: human oversight is mentioned in relation to high-risk systems, but the text provided does not specify which systems fall into this category or what requirements are imposed on administrations that adopt them.
In summary, the picture that emerges is one of a push toward more structured governance of public AI, accompanied however by a possible gap in alignment with the rules on responsibility for measures. The source does not offer elements to establish whether this is a transitional ambiguity or a conflict destined to be clarified by subsequent acts.
