Artificial intelligence: new rules in Italy on civil and criminal liability and facial recognition from 30 September 2026

As of 30 September 2026, Legislative Decree No. 160 of 9 September 2026, published in Official Gazette No. 214 of 15 September 2026, will enter into force in Italy. The decree forms part of the framework established by Regulation (EU) 2024/1689, the so-called AI Act, and governs certain national aspects relating to the use of artificial intelligence, with particular regard to police activities, civil and criminal liability, and the use of biometric identification systems.

The decree implements the delegation granted to the Government by Law No. 132 of 23 September 2025, which provided for measures to align the Italian legal system with the AI Act and for specific rules on the use of artificial intelligence systems in police activities (Article 24 of Law No. 132/2025).

Structure of the decree

The decree consists of 22 articles divided into three Titles:

• Title I governs the use of AI systems in police activities, from testing and personnel training to remote biometric identification;

• Title II introduces criminal and procedural provisions concerning the unlawful use of AI and provides remedies for the compensation of damages suffered by injured parties;

• Title III contains transitional and financial provisions.

The national rules must be read in conjunction with the European regulation, which classifies certain remote biometric identification systems and specific systems used in law-enforcement activities as high-risk systems, insofar as their use is permitted under Union or national law (Annex III to Regulation (EU) 2024/1689).

Criminal-law developments

The decree introduces new Article 437-bis into the Italian Criminal Code. The provision punishes the failure to adopt the technical safety measures or human-oversight measures required for high-risk AI systems where that failure creates a concrete danger to life or to public or individual safety. The penalty is imprisonment for between one and five years; it increases to between two and eight years where the danger concerns national security.

The decree also governs the unlawful alteration of high-risk AI systems, providing for imprisonment of between two and six years, with an increase in the most serious cases.

These provisions supplement the criminal-law amendments already introduced by Law No. 132/2025, including the new general aggravating circumstance concerning offences committed through the use of artificial intelligence systems and the offence of unlawfully disseminating images, videos or voices falsified or altered through AI (Article 26 of Law No. 132/2025).

The catalogue of predicate offences relevant to corporate administrative liability under Legislative Decree No. 231/2001 is also extended. New Article 25-vicies provides, for the offence under Article 437-bis of the Criminal Code, for a financial penalty of between 600 and 1,000 quotas, in addition to the disqualifying sanctions applicable under the Legislative Decree No. 231/2001 system.

Biometric identification and facial recognition

The decree regulates real-time remote biometric identification by law-enforcement authorities, permitting it in the cases provided for by law and subject to prior authorisation by the judicial authority. Its use must nevertheless comply with the conditions, limits and safeguards established by the European regulation and by national data-protection legislation.

The establishment of biometric databases through the mass and indiscriminate collection of information from the web is expressly prohibited.

In the context of video surveillance in places or at events characterised by specific public-order requirements, facial images may be associated with personal data and, where applicable, with an assigned seat, without biometric data being processed at the initial stage. Facial recognition may subsequently be activated after an offence has been committed, subject to the conditions laid down by the decree. The data collected must be automatically deleted after seven days, subject to any exceptions or requirements provided for by law.

The criteria for identifying the places and events concerned are to be established by decree of the Minister of the Interior, after consultation with the Italian Data Protection Authority.

At the procedural level, new Article 359-ter of the Italian Code of Criminal Procedure governs identification and localisation through AI systems intended for real-time remote biometric identification. The measure may be used to confirm the identity of, or search for, persons suspected of certain offences covered by the AI Act and punishable by a maximum term of imprisonment of at least four years, as well as to search for fugitives.

Civil-law developments

On the civil-law side, the decree strengthens the position of persons injured by the use of AI systems through a series of measures designed to reduce the information imbalance between the parties.

The main developments include:

• the right to access the system’s technical documentation, logs and information concerning human oversight; failure to comply with an order to produce such material may allow the court to deem the facts alleged by the applicant to be established;

• a presumption of a causal link between the breach of obligations laid down by the AI Act and the damage suffered, subject to proof to the contrary;

• an alternative forum for consumers, consisting of the court of the injured person’s residence or domicile;

• a direct action against the insurer providing the tortfeasor’s civil-liability cover, within the limits of the policy limit.

Compliance of an AI system with the European regulation, even where formally certified, does not in itself exclude civil liability for the damage caused. The assessment must take account of the operators’ conduct, the applicable obligations, the damage and the causal link.

How to obtain compensation for damage caused by artificial intelligence

Legislative Decree No. 160/2026 introduces specific procedural tools to assist persons seeking compensation for damage attributable to the operation or use of an artificial intelligence system. The rationale is to reduce the information imbalance between the injured party, who will often lack the technical data needed to prove the claim, and the provider or deployer of the system.

An order to produce technical documentation

The injured party may ask the court to order the opposing party or a third party to produce evidence concerning the operation of the AI system. Such an order is not automatic: the court must assess the prima facie plausibility of the compensation claim and the proportionality of the request, including in relation to the protection of trade secrets and confidential information.

Potentially relevant material includes the logs of high-risk systems, documentation relating to the risk-management system and information concerning human oversight. These documents may make it possible to reconstruct the system’s behaviour, any anomalies that occurred and the degree of control exercised by the operators.

Failure to comply, in whole or in part and without justified reason, with an order addressed to a party may allow the court to deem the facts alleged by the applicant to be established. Where the order is addressed to a third party, non-compliance is instead subject to a financial penalty. The rules do not therefore eliminate the need to plead the damage and identify the link between the AI system and the harmful event; they provide the injured party with a means of obtaining information that would otherwise be difficult to access.

Presumption of the causal link

For actions based on a breach of the obligations laid down by the European regulation on artificial intelligence, the decree provides for a rebuttable presumption that a causal link exists between the breach and the damage. The presumption may be rebutted by contrary evidence and does not therefore create an automatic entitlement to compensation.

The injured party must still identify the relevant conduct or omission, establish the existence and extent of the loss, and connect the compensation claim to the AI system used. The presumption operates at the level of causation and does not replace proof of the other elements of liability.

Compliance of the system and civil liability

Compliance of the system with the obligations under Regulation (EU) 2024/1689, even where formally certified, does not in itself exclude the defendant’s civil liability. Certification is one element to be assessed as part of the overall analysis; it does not prevent the court from determining whether damage was caused in the specific case, whether human-oversight obligations were complied with and whether the system was used for its intended purpose and in accordance with the applicable instructions.

Consumer protection and insurance

Where the injured party is a consumer, the proceedings may also be brought before the court of the consumer’s place of residence or domicile, as an exception to the ordinary rules on territorial jurisdiction.

The decree also regulates the relationship with insurers. A person intending to bring a compensation claim may ask the alleged liable party whether it has civil-liability insurance capable of covering the damage alleged. The recipient of the request must communicate, within 30 days, the policy details and the identity of the insurer. Failure to respond, or an incomplete response, may be taken by the court as an adverse evidentiary inference.

The injured party is also granted a direct action against the insurer, within the policy limit. In the relevant proceedings, the insured party must participate as a necessary party. The injured party may therefore proceed directly against the insurer, but the protection remains limited by the existence and scope of the cover, the operation of the policy and the insured limit.

Overall, the new provisions do not establish automatic liability for every instance of damage produced by an AI system. Rather, they strengthen the injured party’s procedural position by facilitating access to technical evidence, easing — in the cases provided for — the burden of proving causation, and making it easier to identify any available insurance cover.

Concluding remarks

Legislative Decree No. 160/2026 represents a comprehensive effort to coordinate the European regulation on artificial intelligence with the Italian legal system. The decree introduces specific rules for police activities, new criminal offences or aggravating circumstances, corporate-liability provisions and forms of civil protection that are more favourable to injured parties.

Companies that develop or deploy high-risk AI systems will therefore need to review their safety protocols, human-oversight procedures, personnel training and technical-documentation practices. Particular attention should also be paid to the organisational safeguards relevant to corporate liability under Legislative Decree No. 231/2001.

The framework must be read together with the AI Act, which requires Member States to establish effective, proportionate and dissuasive penalties for breaches of the regulation (Article 99 of Regulation (EU) 2024/1689).

© Avv. Gabriele Lanzi