The future is already here. And, according to the administrative courts, it may not even require a meaningful “human reserve”.
In judgment no. 758/2026 of 1 June 2026, the Marche Regional Administrative Court addresses, from a highly unusual perspective, the increasingly relevant issue of artificial intelligence in the drafting of administrative acts. The Court reaches a remarkable conclusion: human intervention may be limited to the preliminary decision to use an AI-based procedure, provided that the deployer — whether a public or private entity using AI systems — is able to provide clear and meaningful explanations regarding the role played by AI in the decision-making process and the key elements of the decision adopted.
Even more striking is the Court’s view that an administrative act containing incorrect or even non-existent case law references may still be considered fully lawful.
At the heart of the decision lies the concept of the so-called “human reserve” in administrative decision-making. According to the Court, depending on the type of procedure involved, this requirement may be satisfied even if human intervention is limited to the initial decision to rely on AI, while subsequent oversight focuses only on verifying the proper functioning of the system and ensuring adequate transparency.
This oversight may be further reduced when AI reproduces a deeply human habit: reinforcing administrative decisions with extensive references to case law. This tendency is often linked to the growing influence of so-called living law and to the well-known fear of signing — the tendency of public officials to rely heavily on jurisprudential references in order to reduce perceived personal responsibility.
But what happens if those references are simply hallucinations generated by AI? What if the public official fails to detect them?
According to the Marche Court, nothing.
Even assuming that the case numbers and judicial references cited in the act are inaccurate, and even assuming such inaccuracies derive from the use of AI, the act remains valid so long as the legal principles underlying those references are, in substance, correct and can be found elsewhere in the legal system.
The Court compares this to the use of AI by a lawyer researching case law for litigation. In its view, this is merely a modern evolution of traditional legal research methods — replacing manual consultation of journals and case law collections with digital tools — without affecting the substantive decision-making process.
The ruling therefore appears to open the door to the broader use of AI in public administration, even where outputs are not entirely reliable.
This raises obvious concerns regarding the principle of adequate reasoning of administrative acts under Article 3 of Law No. 241/1990, as well as questions of consistency when compared to the stricter standards applied to lawyers relying on AI-generated legal hallucinations. Recent case law, including Order No. 11431 of 26 March 2026 of the Italian Supreme Court, has taken a far less permissive approach.
The broader question remains unavoidable: are we moving toward a future in which administrative acts drafted by AI — with minimal human oversight — will eventually be reviewed by equally automated systems?
In reflecting on these questions, it is worth recalling the words of Pope Leo XIV in Magnifica Humanitas:
“It is now up to us to confront the challenges of our time with clarity and responsibility. We must adopt adequate regulatory tools capable of protecting justice and containing the distortive effects of technological power.”
The question is no longer whether AI will become part of public decision-making. The real question is under what rules, with what safeguards, and with what remaining space for human judgment.
