2026-09-28 - Legislative Decree 160/2026: biometrics, documentation and liability in the implementation of Italian Law 132/2025
On 15 September 2026,
Legislative Decree no. 160 of 9 September 2026
was published in the Italian Official Gazette, General Series no. 214. The decree aligns national law with
as regards the use of artificial intelligence systems for police activities and civil and criminal liability. It is the first of the two legislative decrees adopted under the delegation laid down in Article 24 of
Law no. 132 of 23 September 2025
. The text contains no specific provision on its entry into force: the ordinary period of fifteen days from publication therefore applies, with effect from 30 September 2026.
A reader stopping at the title might regard it as a sectoral measure addressed to the police forces alone. That would be an incomplete reading. The decree contains provisions that concern anyone who designs, provides or uses artificial intelligence systems in any field: a new criminal offense, the extension of corporate liability and, above all, a set of procedural rules on actions for compensation of damage caused by AI systems, applicable to any system and not only to high-risk ones.
This contribution offers an overview of the final text and opens a new series of in-depth contributions on the implementation of Italian Law 132/2025, which will continue over the coming weeks with examinations of individual aspects.
From the draft to the final text
In this blog, I commented on the
preliminary examination of 10 June 2026
, based on the official press release and the draft versions then circulating. The Council of Ministers gave final approval to the two decrees at its meeting of 4 August 2026, and
(in Italian) reports the changes made following the opinions of the parliamentary committees, the Unified Conference and the Italian Data Protection Authority (Garante per la protezione dei dati personali). For the decree examined here, the changes concern, in short:
- the minimum requirements of the reference database for retrospective facial recognition, with explicit deletion obligations and guarantees that the data used for biometric comparison cannot be incrementally expanded;
- a separate provision on the professional user of high-risk systems who intentionally fails to adopt human oversight measures;
- the participation of local authorities in the installation and maintenance of the AI components of video surveillance systems used to combat crime;
- the clarification of the Garante’s powers, its involvement in regulatory sandboxes and the rules on processing sensitive operational data in research activities.
Other choices were left unchanged. Among them is the attribution to the public prosecutor of the power to authorize real-time remote biometric identification for preventive purposes, an aspect that I flagged in the June contribution as one to be measured against Article 5(3) of the Regulation.
The structure of the decree
DivisionArticlesSubject matter
Title I, Chapter I1 and 2Subject matter, principles and definitions
Title I, Chapter II3 to 6Research, development and use of AI systems in police activities, research collaborations, regulatory sandboxes, staff training
Title I, Chapter III7 to 10Categorisation of biometric data, real-time remote biometric identification, retrospective facial recognition
Title II, Chapter I11 to 15Amendments to the Criminal Code, the Code of Criminal Procedure and its implementing rules, and Legislative Decree 231/2001
Title II, Chapter II16 to 20Civil procedural tools for compensation of damage caused by the use of AI systems
Title III21 and 22Transitional and financial provisions
Artificial intelligence in police activities
Title I rests on a declared principle: artificial intelligence is a support tool, not a substitute for human decision-making. Article 1 specifies that the decree does not entail new obligations compared with those laid down by the Regulation for the systems and models used in police activities. Article 3 requires 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, carried out by staff identified under each police force’s internal procedures and documented to ensure traceability. For high-risk systems, human oversight must be effective and compliant with Article 14 of the Regulation, and the staff in charge must have adequate competence and training.
The decree also regulates research collaborations with universities, research bodies and public and private entities (Article 4), provides the legal basis, under Article 59(2) of the Regulation, for the processing of personal data in regulatory sandboxes (Article 5) and sets the minimum learning outcomes of the training courses for staff (Article 6), implementing Article 4 of the Regulation.
Chapter III is devoted to biometrics: categorization of biometric data, real-time remote biometric identification for preventive purposes and for the search for missing persons or victims of specific offenses, and retrospective facial recognition on video surveillance systems. The common thread is procedural: authorizations, time and territorial limits, comparison databases built for each use, fundamental rights impact assessment and data protection impact assessment, logs that cannot be altered, notification to the Garante, a ban on databases fed through untargeted scraping, and a ban on decisions producing adverse legal effects based solely on the results of facial recognition.
These aspects, which sit at the intersection of the AI Act and data protection law, will be addressed in the next contributions in the series.
Criminal provisions and corporate liability
Title II also contains criminal and criminal procedure provisions worth flagging for anyone working in compliance, although they fall outside the scope of this series. The new Article 437-bis of the Criminal Code addresses failure to adopt the security and human oversight measures required for high-risk AI systems and the alteration of such systems, with a separate provision on the professional user who intentionally fails to adopt human oversight measures. The new Article 25-vicies of Legislative Decree 231/2001 extends the list of predicate offenses for corporate liability, referring to that offense and to the offense under Article 612-quater of the Criminal Code. The new Article 359-ter of the Code of Criminal Procedure governs real-time remote biometric identification in criminal proceedings.
For organizations, the impact is first and foremost organizational: the human oversight required by the AI Act for high-risk systems is not a formality, and organizational, management and control models will need to be reviewed in light of the new provisions.
Civil liability: access to evidence and presumptions
Chapter II of Title II is, in my view, the part likely to have the greatest practical impact because it applies to contractual and non-contractual actions for damage caused by the use of an artificial intelligence system, without being limited to high-risk systems.
The decree introduces a mechanism for access to evidence: the court may order disclosure of evidence relevant to the system’s functioning, including logs, documentation of the risk management system, technical documentation, and information on human oversight required by the Regulation. It also provides for a rebuttable presumption of causation where the damage results from a breach of obligations under the Regulation applicable to the case at hand, an alternative forum for injured parties acting for purposes outside their professional activity, and a direct action against the insurer covering the defendant’s civil liability.
It also states a principle that deserves emphasis: compliance of the system with the obligations of the Regulation, even where certified, does not in itself exclude the defendant’s liability.
As a result, the documentation required by the AI Act ceases to be merely a requirement towards the authorities and becomes a central element in litigation. This is a ground the Union left uncovered after withdrawing the proposed AI Liability Directive, as recalled in the June contribution: the national choice deserves dedicated analysis, which will be the subject of one of the forthcoming articles.
Entry into force and transitional regime
The decree enters into force on 30 September 2026. However, three timing aspects need to be distinguished.
The first concerns existing systems. AI systems which, on the date of entry into force of the decree, are already in use in police activities, under development or covered by contractual relationships must be made compatible with the provisions of Chapter II of Title I within one year.
The second concerns the provisions of Title I whose application is laid down directly by the AI Act or depends on it: for these,
(in Italian) refers to the dates of entry into force and full implementation of the Regulation. That reference must be read in light of the timetable amended by
, examined in the
contribution on its publication
: Sections 1, 2 and 3 of Chapter III of the AI Act, except Article 6(5), which include, among other things, the requirements for high risk systems, human oversight, logging, deployers’ obligations and the fundamental rights impact assessment, apply from 2 December 2027 to Annex III systems and from 2 August 2028 to Annex I systems, while the original prohibitions in Article 5 have applied since 2 February 2025.
The third concerns Title II, to which the reference in Article 21(2) does not apply. However, where a provision presupposes a breach of an obligation under the AI Act, it will be necessary to verify that the obligation applies to the case at hand.
The decree includes a financial neutrality clause: the administrations concerned act within the resources available under current legislation. This is the aspect on which, in the June contribution, I recalled the requirement of adequate resources laid down in Article 70(3) of the Regulation.
The second implementing decree
At the same meeting of 4 August, the Council of Ministers also gave final approval to the second decree, concerning the powers of the national authorities and the use of artificial intelligence in education and training. According to the press release, the text defines the national governance set-up, the penalty framework, the operation of regulatory sandboxes, and the use of AI in education, the professions, employment, healthcare, and public administration.
I will address it once it is published in the Official Gazette, not least because some topics, starting with training and the allocation of competences among authorities, need to be read across both texts.
The next contributions in the series
The series on Legislative Decree no. 160/2026 will continue with four in-depth contributions:
- facial recognition and remote biometric identification, read in light of Article 5 of the AI Act;
- impact assessments, logs and documented human review, and what organizations can draw from them;
- research collaborations, regulatory sandboxes and ownership of trained models;
- the documentation required by the AI Act in litigation for damages.
For those wishing to navigate the regulatory framework referred to in this contribution, 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:
Legislative Decree no. 160 of 9 September 2026
and
Law no. 132 of 23 September 2025
in the Official Gazette,
press release no. 185 of the Council of Ministers
on the Government’s website (in Italian), and
and
on EUR-Lex.
New to Gemini? This capsule is best read with a Gemini client, such as Lagrange.