Damages related to the use of AI: Italy allows direct lawsuits against the insurer. What’s next?

The new Article 20 of Legislative Decree No. 160/2026 allows those who have suffered damages from the “use” of AI systems to bring a direct lawsuit against the third-party liability insurer of the alleged damaging party. The liability regime applicable to the insured does not change on the merits. However, there might be several effects on insurers: greater exposure to litigation and related costs, a rethink of policy wordings, and a more careful analysis at underwriting stage.

Similarly to the models already tested for third-party liability in the motor and medical liability sectors, Article 20 of Legislative Decree No. 160/2026 (Decree) has introduced, effective from 30 September 2026, the possibility for the damaged party to bring a direct lawsuit for damages arising from the use of artificial intelligence (AI) systems against the insurer covering the civil liability of the party deemed liable, with the insured/alleged damaging party remaining a necessary party as co-defendant in such proceedings.

The most disruptive impact of the new provision lies in the broad and generic scope of the term “use” (that is not defined in the Decree itself, nor in the AI Act, from which the Decree borrows other definitions) Such term might, therefore, abstractly encompass a very wide range of potentially liable players, acting with different roles along the AI systems supply chain. These include the provider, the developer, the distributor, the deployer, the professional who relies on the output, and the person in charge of human oversight. Damages may, in fact, arise from the dataset used by the AI system, improper use of the system, design errors, the configuration adopted by the user, an incorrect or omitted update, insufficient instructions, or human reliance on an output that was not adequately verified.

This will inevitably mean that a wide and heterogeneous types of insurance policies may also be triggered, as the risk could fall within one or more covers offered under general liability, professional indemnity, product liability, tech E&O and/or cyber liability policies, and that several insurers may as well be involved, given the possible concurrent liability of different AI supply chain players for the same damage.

The liability regime applicable to the party who “uses” the AI system that caused the damage does not change on the merits. This assessment – that remains the necessary prerequisite for any possible judgment against the liability insurer of the damaging party – will be carried out under the ordinary civil liability rules and principles, establishing first of all the causation link with respect to each player of the AI supply chain, and without prejudice to the liability regime provided by the legislation transposing Directive (EU) 2024/2853 on defective products liability, as expressly mentioned in the Decree. The Decree introduced, however, a few new provisions from an evidentiary perspective: on one hand, the Court has the power to order the alleged damaging party to disclose all the evidence on how the AI system works; on the other hand, subject to some specific conditions, there is a presumption of the existence of a causal link between the breach of one or more obligations imposed by the AI Act and the damage, that can be rebutted by the damaging party through adequate evidence proving the contrary. The Decree also clarifies that the compliance of the system involved with the AI Act requirements does not necessarily per se rule out the liability.  by verified.

What changes from an insurance perspective is the procedural order in the relationship between insured and insurer. Before this new legislation, in fact, the damaged party would have sued the alleged damaging party, who would then have most likely joined as a third party into the proceeding their own insurer seeking indemnity. Now, the damaged party may sue directly the insurer of the damaging party, the latter being in any case summoned as well as a necessary co-defendant, since the assessment of its actual liability remains the fundamental prerequisite for any judgment against the insurer. This will most likely result in a broader litigations for insurers, as the decision of their involvement would no longer be made by the insured who, with the support of its broker, would normally make a preliminary analysis to identify whether and which insurance cover might be relevant in the specific case, but would be made by the damaged party, who does not know anything about the actual content of the damaging party’s insurance contracts. This means that the damaged party might, therefore, decide to sue all the insurers of all potentially liable parties under all the policies in place, just to increase the chances of success in the lawsuit, or to secure a broader pool of “qualified” counterparties in a possible settlement scenario.

From a practical standpoint, Article 20 of the Decree provides that the damaged party may first ask the alleged liable party whether it holds any third-party liability policies covering the AI related damages and request the details of the relevant insurance contract. The damaging party must respond within thirty days. If the damaging party does not respond or submit a partial/incomplete response, this might be considered by the Court as circumstantial evidence pursuant to Article 116 of the Italian Code of Civil Procedure. This is a mere option for the damaged party and does not represent a mandatory preliminary step for the direct lawsuit against the insurer, that can be brought even without this information. Still, it remains unclear, except where third-party liability cover is compulsory and the details of the relevant insurance contract are public (as for lawyers PI policies), how the damaged party, lacking information from the damaging party, could identify the insurer to sue directly and the policy to be invoked in court.

The most immediate consequence of this potential risk of increased litigation against insurers is “just” an increase in defense costs, as insurers may be forced to bear unnecessary legal expenses to defend proceedings in which their policies do not come into play at all.

Once the direct lawsuit has been brought against the insurer, in fact, still it will be necessary to assess, first of all, the liability of the insured and, only then, the assessment will move into insurance coverage, that will be made as usual: it will be assessed whether the event giving rise to the damage claimed by the claimant falls within the scope of the insured risk; the insurer may raise against the damaged party all defenses arising from the insurance contract (and, as expressly provided by the Decree, based on facts occurred before the claim, which is of little practical impact if we consider that most coverage defenses are based on prior facts), including the applicable limits of indemnity.  The Decree further clarifies that the insurer, once made a payment, will have a recourse right against the insured, to the extent that it would have had a contractual basis to refuse or reduce its obligations.

It will, therefore, become material for insurers to understand and decide which approach to adopt towards AI risk, which already latently exists as silent AI under existing policies. The options are twofold: exclude it, or cover it and expressly governing it, maybe by introducing specific sub-limits, as no statistics are currently available for this type of risk which could potentially give rise to high exposures. Introducing specific exclusions for AI damage may not always be a feasible option in the current soft market characterized by pressing demand from brokers and insureds for this type of cover, which is considered essential now that AI is everywhere.

The efforts of insurers opting for affirmative cover will need to focus, first of all, on policy wordings, paying careful attention to definitions and to the triggering events for each cover considered, to related and/or multiple claims clauses (as an error embedded in a model or an update may produce numerous harmful outputs, affecting different parties and generating different claims) and “Other insurance" clauses (as several policies could in theory be triggered by the same loss).

Attention must, then, be paid also at the underwriting stage, by introducing into the proposal forms specific questions aimed at determining the insured's exposure to AI risk. More in detail, the underwriters should be able to understand and assess: which AI systems the insured uses; how the insured implements and uses them; the insured's compliance with AI legislation, including with any disclosure obligations when using AI systems; the insured's position in the AI supply chain, and any reliance on external AI providers, including their respective contractual rights and obligations; the insured's ability to retain technical evidence on how the AI systems work.