Justice and AI

No to the anonymisation of sentences without human control

Lazio Regional Administrative Court rejects the use of automatic systems: there is a risk that the algorithm will turn from a transparency tool into an opacity mechanism

3' min read

3' min read

Among the most pervasive applications of artificial intelligence in judicial systems, the anonymisation of decisions represents, at least on the surface, a virtuous intervention: aimed at protecting the confidentiality of the parties involved, without compromising the principle of publicity of judicial decisions. However, the Italian case recently dealt with by the Lazio Regional Administrative Court (judgement no. 7625 of 17 April 2025) shows how, in the absence of a careful and conscious balancing of rights, the algorithm can turn from an instrument of transparency into a mechanism of opacity.

The case

.

The affair is, in fact, emblematic. With the replacement of the National Jurisprudential Archive (Agn) with a new public database (Bdp), the Ministry of Justice intended to make the patrimony of Italian judgments more accessible. But when the new portal opened, legal practitioners - lawyers and scholars - found themselves faced with documents emptied of their informative content: removed were not only the names of the parties, but also dates, normative and jurisprudential references, even essential factual coordinates.

Loading...

Aradical automation of the anonymisation process, devoid of human control and disengaged from context, has produced the paradoxical effect of annihilating the cognitive, orientative and guarantor function of jurisprudence. In the name of privacy it has ended up denying the right to knowledge of living law, calling into question the very intelligibility of the legal system.

The Tar's decision

.

The Regional Administrative Court, in a decision destined to set the standard, has made it clear that anonymisation cannot be reduced to a mere technical operation, nor flattened on a blind automatism. On the contrary, it affects the quality of justice and the effectiveness of the right of defence. It betrays the founding principle of law - da mihi factum, dabo tibi ius - any practice that obliterates facts, making it impossible to apply the rule consistently to the concrete case.

The ruling also highlights how generalised and indiscriminate anonymisation conflicts with constitutional and supranational principles. Article 6 of the European Convention on Human Rights and Article 47 of the Charter of Fundamental Rights of the European Union - as well as Article 111 of the Italian Constitution - enshrine not only the publicity of hearings, but also the knowability of decisions. Publicity that does not end with the formal visibility of the text, but requires that the judgment be comprehensible, in its logic, in its motivational path, which cannot disregard the concrete case, and in its systemic implications.

AI governance

The critical point, however, goes beyond the specific case. It is structural. The European AI Act, in its current wording, expressly excludes systems for the automatic anonymisation of judicial decisions from the category of high-risk systems. A choice that, in the light of what has happened, turns out to be questionable. If Recital 61 qualifies these applications as mere 'ancillary administrative activities', with no direct impact on the administration of justice, the Italian case shows that the outcome of their adoption may be anything but neutral.

It is not a matter of invoking technological prohibitionism, but rather of introducing - as in other areas of AI governance - actual forms of algorithmic accountability. Human oversight, the right to explain, traceability of changes, and impact analysis on fundamental rights must be indispensable elements, even for apparently 'technical' solutions such as anonymisation.

The risk we run is that of a faceless justice, without memory and without orientation. A 'closed' justice, more like a black box than a public service based on transparency and trust. And while it is true that every fundamental right requires a balancing act - and privacy is no exception - it is equally true that this cannot be done automatically, impersonally and unreflectively. Because justice, in a constitutional democracy, cannot only be rendered: it must also be visible, understandable and controllable by everyone.

In the end, as the ruling warns, it is not the technology that needs to be demonised, but its uncritical use. If we leave it to algorithms alone to decide what can be known and what must be hidden, the risk is not that of protecting too much privacy. It is that, far more serious, of erasing justice.

Copyright reserved ©
Loading...

Brand connect

Loading...

Newsletter

Notizie e approfondimenti sugli avvenimenti politici, economici e finanziari.

Iscriviti