2026-10-05 - FRIA, DPIA and logs: what Italian Legislative Decree 160/2026 requires and what organisations can learn from it


This contribution continues the series devoted to Italian

(in Italian), after the

and the analysis of

. The subject here is the set of obligations accompanying the use of artificial intelligence systems in police activities: impact assessments, logs, notifications, and human review.

This is where the decree most closely intertwines

with the data protection rules applicable to processing for police purposes, namely

(in Italian), which transposes

. It is also where private organizations can draw the most useful lessons, even though they are not its addressees.

Overview of the obligations

ObligationLeg. Decree 160/2026AI ActData protection

Qualified and traceable human reviewArt. 3(4)Art. 14; Art. 26(2)Art. 8 Leg. Decree 51/2018

Effective human oversight of high risk systemsArt. 3(5)Art. 14

Fundamental rights impact assessmentArt. 9(1)Art. 5(2); Art. 27

Data protection impact assessmentArt. 9(1); Art. 10(6)Art. 27(4)Art. 23 Leg. Decree 51/2018

Prior consultation of the GaranteArt. 10(6)Art. 24 Leg. Decree 51/2018

Logs that cannot be altered, kept for five yearsArt. 9(2); Art. 10(8)Art. 12(3)Other retention periods under Art. 10 of Presidential Decree 15/2018 continue to apply

Notification of each use to the GaranteArt. 9(4)Art. 5(4)

Technical requirements and bias monitoringArt. 9(5) (ministerial decree)Arts. 10 and 15

The references to the AI Act must be read in light of the timetable amended by

: Articles 10, 12, 14 and 15, in Section 2 of Chapter III, and Articles 26 and 27, in Section 3, apply to Annex III systems from 2 December 2027, while the provisions of Article 5 referred to here already apply. The link with national dates runs through the reference contained in Article 21(2) of the decree.

Human review: qualified, documented, traceable

The first obligation concerns neither biometric systems alone nor high-risk systems alone. Article 3(4) provides that the use of AI systems and models in police activities must include adequate forms of qualified human review of the results of automated processing before they are used in acts and measures affecting the legal sphere of the persons concerned. Staff identified under each police force’s internal procedures carry out the review, and it is documented to ensure traceability.

For high-risk systems, paragraph 5 adds that human oversight must be effective and compliant with Article 14 of the Regulation, and that the staff in charge must have the necessary competence and training.

The distinction between the two paragraphs is significant. Human oversight under Article 14 of the AI Act is a system requirement that the provider must enable and the deployer must implement. The review under Article 3(4), by contrast, is a procedural step on the output, which precedes its entry into a legal act and must leave a verifiable trace. In my view, it is the most concrete translation of the principle that artificial intelligence supports rather than replaces human decision-making: the principle becomes demonstrable only if the review is documented.

FRIA and DPIA: distinct but coordinated assessments

The distinction between the two assessments was examined in the contribution

. Coordination can avoid duplication and allow relevant analyses to be reused, but it must preserve the recognisability of the conditions, contents and outcomes of each assessment.

In the context examined here, the applicable rules must also be distinguished: for processing carried out by the competent authorities for the police purposes covered by the decree, the data protection impact assessment is governed by Legislative Decree 51/2018. The earlier contribution therefore provides the methodological framework to be applied within the specific regime of Legislative Decree 160/2026.

For real-time remote biometric identification, both in the prevention track and in criminal proceedings, Article 9 requires the controller to complete in advance the fundamental rights impact assessment provided for in Article 27 of the AI Act and to carry out the data protection impact assessment provided for in Article 23 of Legislative Decree no. 51/2018, with the prior consultation governed by Article 24. For retrospective facial recognition, Article 10(6) requires a data protection impact assessment and consultation of the Garante, allowing a single assessment for all similar systems.

The two assessments have partly overlapping but not identical subjects. The data protection impact assessment addresses the risks arising from the processing of personal data; the fundamental rights impact assessment has a different subject, centered on the impact of the use of the system on fundamental rights in the specific context, and covers the processes in which the system is used, the categories of persons affected, the specific risks of harm, the human oversight measures, and the measures to be taken if the risks materialize. Article 27(4) of the AI Act, as amended by Regulation (EU) 2026/1744, allows the fundamental rights impact assessment to include cross-references to the relevant sections of the data protection impact assessment or to incorporate parts of it, and paragraph 5 tasks the AI Office with developing a questionnaire template, including through an automated tool. Article 27 will apply to Annex III systems from 2 December 2027; for real-time remote biometric identification, however, the fundamental rights impact assessment is already a condition of use under Article 5(2).

The decree does not regulate the operational coordination between the two assessments, which is left to the controller. A coordinated process that allows common analyses to be reused while keeping the conditions, contents, and outcomes of each assessment recognizable is an organizational choice consistent with the Regulation. What matters is consistency between the two assessments, which different authorities will check and, as a later contribution will show, may also become relevant in court.

Logs: what, for how long, for whom

Every use of real-time remote biometric identification systems is automatically recorded in logs that cannot be altered and that must include the data required by Article 12(3) of the Regulation: the period of each use, the reference database, the input data that led to a match and the identification of the persons who verified the results. The same applies to access to and operations on retrospective facial recognition systems.

The logs are kept for five years and are accessible only to the authorities competent to verify the lawfulness of processing, for internal control and in criminal proceedings. The retention periods laid down for police purposes by Article 10 of

(in Italian) continue to apply to the other personal data processed.

This point deserves emphasis, because the decree distinguishes different retention regimes. Comparison databases are deleted when the authorization expires; images stored at public venues and events are deleted after seven days; the records of operations, which themselves contain personal data, such as the inputs that led to a match and the identification of those who verified the results, are kept for five years; other data processed follow the periods laid down for police purposes. Deleting the comparison set therefore does not erase every trace of the processing: retaining the records makes subsequent review possible.

Notification to the Garante

After each use of real-time remote biometric identification systems, the controller notifies the Garante of the essential information on the authorization granted and on the result obtained, without including sensitive operational data, in accordance with Article 5(4) of the AI Act. Notification is subject to prior clearance by the competent judicial authority, which may defer it for up to three months, renewable once, where specific confidentiality requirements apply. For multiple activations in the same operational context or for homogeneous uses, notification may be made periodically and cumulatively, the requirement of prior clearance remaining unaffected.

The choice of the Garante as the single addressee can be understood in light of the second decree implementing Law no. 132/2025, which, according to

(in Italian), confers on the Garante competences over systems used in law enforcement. I will address it once the text is published.

The ministerial decree: accuracy, bias, security

Much of this framework’s effectiveness depends on a decree of the Minister of the Interior, to be adopted in agreement with the Minister of Justice after consulting the Garante and the national AI authorities. Article 9(5) lists its contents: minimum requirements of reliability and accuracy of the systems; arrangements for periodic performance monitoring, with particular regard to detecting and mitigating any bias or discriminatory effects; minimum requirements of the reference databases, including security, deletion times and methods and guarantees against incremental expansion; security measures and measures preventing unlawful access; and the information to be attached to the authorization request, including the technical characteristics of the system, the results of accuracy checks and the error management mechanisms.

The last point is, in my view, the most relevant: the authority called upon to authorize does not decide only on the necessity and proportionality of use, but also based on technical information on the system’s reliability. This strengthens the quality of prior review, provided the information is understandable to those who must assess it.

What organizations can learn

The decree concerns the police forces, but the model it builds offers useful lessons for any organization using AI systems, particularly high-risk ones, that is required to carry out a data protection impact assessment under Article 35 of

or, where applicable, a fundamental rights impact assessment under Article 27 of the AI Act.

Three in particular:

This last aspect takes on further importance in light of Chapter II of Title II of the decree, which turns the documentation required by the AI Act into evidence in damage claims. That will be the subject of the last contribution in the series.

The next one, instead, focuses on research collaborations, regulatory sandboxes, and ownership of trained models.

For those wishing to navigate the regulatory framework referred to, the

of this site are available: the

, with the state of application of the AI Act, the map of Annex III and operational checklists; the

, devoted to roles, obligations and the chain of responsibility; and the

, on the operational intersections between the two regulations.

The reference sources are available on the institutional websites:

in the Official Gazette,

and

on Normattiva (all in Italian),

,

,

and

on EUR-Lex, and

on the Government’s website (in Italian).


New to Gemini? This capsule is best read with a Gemini client, such as Lagrange.