Artificial intelligence for the police, the Government approves the decrees. Piantedosi: “It won’t be Big Brother”

When a camera looks at a square, the scene still looks familiar. A pole, a lens, some cables, the usual piece of the city that ends up inside a screen. When theartificial intelligenceHowever, the matter changes weight. The gaze becomes calculation, comparison, recognition, archiving. And at that point the question stops being just about safety: it’s about the type of public space we want to live in.

The Council of Ministers has approved in preliminary examination two legislative decrees to adapt Italian legislation to European regulations on AI. One of the most delicate steps concerns the use of systems artificial intelligence by the police forcealso for video surveillance activities, facial recognition and processing of biometric data linked to identification after the commission of crimes.

The Minister of the Interior Matteo Piantedosi presented the provision as a tool to support police action, with final decisions always entrusted to human beings. The formula is reassuring, as often happens when technology enters the halls of power with the word “efficiency” pinned to its chest, a Terminator-like half-smile and little desire to dwell too much on surveillance.

The machine suggests

The decree, as it was presented, is based on a very clear principle: AI can help, accelerate, cross-reference data, report useful elements, but the final decision must remain human. Any use for security purposes should be subjected to qualified human review and surveillancewith protection of personal and sensitive data. The minister ruled out mass surveillance systems and spoke of a ban on large generalized biometric databases.

On paper it all sounds neat. The problem, as always, comes when paper meets practice. A decision can remain formally human even when it arises from an algorithmic suggestion treated as if it were neutral, rapid, objective. Those who work under pressure, with few staff and many cases to manage, can end up relying on the machine much more than the internal regulations allow. The algorithm does not sign, but it can push the hand.

The distinction between support and replacement is subtle. In police matters it becomes even more subtle, because an error does not just produce incorrect advertising or poorly suggested content. It can weigh on controls, suspicions, investigations, personal freedoms. For this reason, guarantees cannot remain a kind of institutional cover slipped over a very powerful technical system. We need legible procedures, traceable responsibilities, external checks, serious training, controlled access, clear retention times. All unspectacular things. In fact, they are the decisive ones.

The face in the system

The most sensitive part concerns thebiometric identification. The provision provides for a double possibility: ex ante use, i.e. before the commission of crimes, in case of danger or threat linked for example to terrorism, in the search for missing persons or victims of trafficking; and use ex post, after the crime, especially to ascertain identities through video surveillance, facial recognition and biometric data. For preventive use, a request from the police commissioner, authorization from the judicial authority, impact assessment on fundamental rights and notification to the Privacy Guarantor are indicated.

Here the language already does half the work: “ex ante”, “ex post”, “guarantees”, “support”, “human review”. Everything seems technical, tidy, reassuring. But then there is a person who crosses a station, a square, a street, and can end up inside a system capable of comparing his face with other data. And then the question becomes less abstract. Much less. Bureaucracy has this gift: it manages to make even a hot issue seem lukewarm.

According to what was illustrated by the Minister of the Interior, i biometric data they will only be kept for seven days and automatically deleted, while the operation logs will remain for five years to avoid abuse. Also indicated is the ban on making decisions about a person based solely on the result of facial recognition and the ban on generalized and non-targeted biometric identification, not linked to criminal proceedings. These are important stakes, of course. A stake remains useful only if someone checks every day that it is not moved a few centimeters.

Europe sets the limits

The European framework, at least on this terrain, does not leave a completely free hand to the States. The AI ​​Act generally prohibits the use of “real-time” remote biometric identification systems in publicly accessible spaces for law enforcement purposes, subject to limited exceptions and strict conditions. The regulation talks about strict necessity, proportionality, prior authorization and precise limits on time, place and purpose.

Italy moves within a European framework that tries to prevent the transformation of public space into a permanent laboratory of automatic recognition. The Government, as expected, presents the measure as a balance between innovation and protection. The real question concerns the concrete application: how many times will this possibility be used? With what controls? From which offices? With what tools? On which databases? With what margins of dispute for those who end up with a wrong identification?

Reducing everything to those for or against technology would be convenient, and also quite useless. No one wants to hinder the search for a missing person or the prevention of an attack. The risk arises when the exception begins to accustom the institutions to an instrument and the instrument begins to ask for more space. The difference lies in the dosage, in the transparency, in the independent control, in the possibility of knowing how those systems are used and how errors are corrected.

The decree must be read in its details and subsequent steps. For now we know that AI is entering the work of the police force in a more structured way, with a series of announced guarantees and a very reassuring lexicon. The guarantees are good, the reference to human decision is good, the declared limits are good. The fact remains that any control technology, once installed, tends to seem indispensable. And when a door opens onto the surveillanceclosing it becomes more and more tiring. Even if they wrote “innovation” at the entrance.