Our Expert in Italy
AI patents italy sits at the centre of one of the most contested areas of intellectual property law as 2026 opens, with generative models, machine-learning-designed compounds and automated engineering tools now embedded in corporate research pipelines. In-house counsel, R&D leaders and patent attorneys face a pressing question: when an artificial intelligence system contributes materially to an invention, who can be named as inventor, who owns the resulting rights, and how will Italian courts and the European Patent Office (EPO) treat validity and enforcement? This guide synthesises EPO practice, the Italian Industrial Property Code and national enforcement procedure into a single practitioner-focused resource.
It is written for teams that need to make filing, contracting and litigation decisions now, not after the law settles.
Quick summary. This guide is for in-house counsel, patent attorneys and R&D leaders. You will learn whether AI can be named as an inventor, how ownership of AI-generated inventions is allocated in Italy, how patentability is assessed, and the evidence and enforcement steps that protect your position. Immediate action: review your AI contracts and put an evidence-preservation protocol in place.
The legal position on AI patents italy can be distilled into a small number of firm points and several areas of genuine uncertainty. Practitioners need to separate what is settled from what remains contested, because the two demand different strategies, settled rules can be relied upon in drafting, while contested points call for defensive documentation and careful contractual allocation.
Three immediate actions follow from these points: audit every agreement under which AI tools are used in R&D to confirm that IP assignment is watertight; implement a documented preservation protocol for models, data and development logs; and draft claims to foreground the human-identified technical contribution rather than the automated process that produced it.
Two bodies of law govern AI patents italy in parallel. Patentability and the designation of inventors on European applications are governed by EPO practice under the European Patent Convention, while ownership, assignment and national enforcement are governed by the Italian Industrial Property Code (Legislative Decree No. 30 of 10 February 2005, as amended). It is essential to keep two distinct questions apart: who is named as inventor is a question of formal entitlement and factual contribution, whereas who owns the patent rights is a question of legal relationship, contract and statutory presumption. A natural person may be correctly named as inventor while a company holds all economic rights by assignment or operation of law.
The EPO has taken a clear position that the designated inventor on a European patent application must be a human being. In the applications that attempted to name an AI system as inventor, the so-called “DABUS” applications, the EPO refused the designations on the ground that the inventor must be a person with legal capacity, a conclusion addressed in its Boards of Appeal practice. The reasoning is formal as much as substantive: the inventor’s designation carries legal consequences, the right to be mentioned, the starting point for entitlement, that only a natural person can bear.
For practitioners, the practical takeaway from current EPO guidance and case law is unambiguous: never attempt to designate a machine as inventor, and always identify the natural person or persons whose contribution justifies the designation.
The EPO has also addressed, through its examination guidance, how AI-assisted and computer-implemented inventions are assessed for patentability. The core approach remains the problem-solution test, with the examiner asking whether the claimed subject-matter makes a technical contribution over the prior art. AI does not receive special treatment; the question is always whether there is a technical effect that a skilled person would not have arrived at obviously.
At national level, the Italian Industrial Property Code governs ownership, assignment and the rights of employee-inventors, and it is the framework a claimant invokes when enforcing a patent before the Italian courts. The Code sets out how rights in service inventions are allocated between employer and employee and the circumstances in which an employee may be entitled to remuneration. It also governs the validity and recordal of assignments, which is where most AI ownership questions are ultimately resolved. The Italian Patent and Trademark Office (UIBM, Ufficio Italiano Brevetti e Marchi), operating under the Ministry responsible for enterprise and economic matters, administers national filing and assignment procedures.
Because specialised business court divisions (sezioni specializzate in materia di impresa) hear patent disputes in Italy, practitioners enforcing AI patents italy must combine EPO-level patentability analysis with Italian procedural strategy, a combination that neither pan-European guidance nor generic commercial commentary provides on its own.
Can an AI be named as an inventor? No. Under current EPO practice and Italian national practice, the inventor named on a patent application must be a natural person. An AI system, however sophisticated, cannot hold the legal status of inventor. Where AI has contributed to an invention, the correct approach is to identify and name the human being who conceived or directed the inventive technical contribution, and to document that contribution clearly.
The direct answer to the most frequently asked question about AI patents italy is that an artificial intelligence cannot be an inventor. Both EPO practice and Italian national patent law proceed from the premise that inventorship is a human attribute. This is not merely a formality that can be corrected later; it shapes how applications must be prepared and how contributions must be documented from the first disclosure. The policy debate about whether the law should recognise machine inventors has been active across Europe and internationally, with WIPO convening discussions on the question, but the operative law as it stands requires a natural person.
Attempting to designate an AI system as inventor produces concrete and avoidable harm. At the EPO, such a designation will be met with a formal objection and, if not corrected, can lead to the refusal of the application. The time lost to correcting a defective designation can jeopardise priority strategies and expose an application to third-party intervention. At national level, a flawed inventorship record can become a line of attack in later litigation, where an opponent may argue that entitlement was never properly established. The safe course is to treat AI as a sophisticated tool and to name the human contributors from the outset, exactly as one would for any other computer-assisted research.
When an AI system plays a substantial role, generating candidate structures, proposing design parameters, or identifying non-obvious combinations, the practical challenge is to identify the human whose intellectual contribution underpins the invention. This is usually the person who framed the problem, selected and configured the model, curated the inputs, recognised the significance of the output, and validated it as a solution. Good practice is to record these human contributions contemporaneously: who defined the research objective, who directed the tool, and who evaluated and selected the result. Contemporaneous invention-disclosure records that capture this human role are invaluable both in prosecution and in any subsequent entitlement dispute, and they are a central element of litigation readiness for AI patents italy.
Once the inventor is correctly identified as a natural person, ownership of the resulting rights is a separate question answered by the legal relationship between the parties and the contracts that govern it. Several recurring scenarios arise in practice, and each carries different default rules and different drafting priorities. Getting ownership of AI inventions right at the contracting stage is far cheaper than litigating it later.
Where an employee creates an invention using an employer’s AI tools in the course of their duties, the Italian Industrial Property Code governs the allocation of rights between employer and employee. The Code distinguishes between inventions made in the performance of an employment relationship and other categories, and it addresses the employee’s entitlement to be recognised as inventor and, in defined circumstances, to remuneration or a fair reward. For AI-assisted R&D, employers should ensure that employment contracts contain clear IP assignment clauses, mandatory invention-disclosure obligations and provisions dealing with the output of AI tools, so that there is no gap between the statutory framework and the company’s commercial expectations.
Many AI inventions arise outside the simple employer-employee relationship. A client may commission an external developer to build models and code, or may use a third-party AI platform on a software-as-a-service basis. In these cases, ownership of AI inventions turns entirely on the contract and, for hosted platforms, on the provider’s terms and conditions. A developer who is not properly bound by an assignment clause may retain rights in the deliverables. A SaaS provider’s standard terms may purport to allocate rights in outputs in ways that surprise the user. The drafting priorities are a clear assignment or licence, express treatment of deliverables and derived IP, and a careful review of any platform terms that touch on ownership of generated output.
Where several people contribute to an AI-assisted invention, a data scientist, a domain expert and a validating engineer, for example, the invention may be jointly owned. Joint ownership raises questions about who may exploit or license the patent and on what terms. These issues are best resolved in advance by a co-ownership or joint-development agreement that specifies the ownership split, licensing rights and decision-making on enforcement, rather than left to default rules that may not suit the commercial reality.
| Scenario | Who is typically named as inventor | Who typically owns initial rights | Key contractual safeguards |
|---|---|---|---|
| Employee uses employer-owned AI in R&D | Employee (natural person) | Employer (under employee invention rules / contract) | Employment clauses, assignment, invention disclosure |
| External AI developer (models & code) used by client | Client researcher (natural person) | Depends: contractual assignment or developer retains rights | Clear licence/assignment, deliverables IP clauses |
| Black-box generative AI (third-party SaaS) output | Human who directed & validated output | Likely user if T&Cs assign rights; otherwise provider | Review provider T&Cs, include bespoke assignment clauses |
| Fully automated system without human oversight | Unclear, risk of challenge; human supervisor typically named | Likely assigned by contract to commissioning party | Contractual assignment, documented human contribution |
Who owns a patent for an AI-generated invention in Italy? Ownership depends on the legal relationship and the governing contracts. For employee inventions, the Italian Industrial Property Code allocates rights between employer and employee. For commissioned development or third-party platforms, ownership follows the assignment, licence or terms of service. Without a clear contractual assignment, a developer or provider may retain rights the commissioning party assumed were its own.
Patentability is where many AI patents italy questions are won or lost. The involvement of an AI system does not change the substantive requirements, novelty, inventive step, industrial applicability and sufficiency of disclosure, but it does change the evidential and drafting challenges in meeting them. Each requirement deserves separate treatment because AI involvement affects them differently.
Novelty asks whether the claimed invention has been made available to the public before the filing or priority date. For AI-generated outputs this is usually a conventional analysis: if the specific technical solution is not disclosed in the prior art, it is novel. The complication is that AI systems trained on large corpora may reproduce or closely approximate existing disclosures, so applicants should conduct careful prior-art searching before filing to confirm that the AI output is genuinely new rather than a near-reproduction of training material.
Inventive step is the most demanding hurdle for AI-assisted inventions. The EPO applies the problem-solution approach: identifying the closest prior art, formulating the objective technical problem, and asking whether the claimed solution would have been obvious to the skilled person. AI involvement raises several distinct issues. First, where the output emerges from a model trained on prior data, an examiner may question whether the solution represents a genuine inventive contribution or merely the predictable product of applying a known technique to a known dataset. Second, the plausibility of the claimed technical effect must be supported in the application, unsupported assertions that a generated compound or design achieves a particular result will not survive scrutiny.
Third, selection from a large space of AI-proposed candidates raises the question of whether the selection itself was inventive or arbitrary.
In practice, the EPO’s examiners expect applicants to articulate the technical contribution in human terms: what technical problem is solved, what technical effect is achieved, and why the skilled person would not have arrived at it obviously. The fact that a model generated candidates does not supply inventive step; the inventive step must reside in the identified technical solution and its demonstrated effect. For AI patents italy, drafting that foregrounds this technical contribution, rather than the automated process, is the single most effective way to defend inventive step both in prosecution and in later litigation.
Sufficiency requires that the application disclose the invention clearly and completely enough for the skilled person to reproduce it. This is acutely difficult for AI-assisted inventions. If the claimed invention depends on a particular model, training dataset or configuration, the applicant may need to disclose enough about the algorithm, parameters or data to enable reproduction. Where the invention is a product or process defined independently of the AI tool, a specific compound, structure or method, sufficiency is more straightforward, because the skilled person can reproduce the result without the model.
The strategic lesson is to claim the technical result in a way that can be enabled by its own disclosure, rather than in a way that depends on reproducing an opaque model.
Three recurring tensions deserve attention. Training data may itself be proprietary or confidential, creating a conflict between the disclosure obligation and the desire to protect trade secrets. The enabling software may be something an applicant wishes to keep secret rather than disclose. And the characterisation of AI as a tool, rather than as an inventor, must be maintained consistently throughout the application, because inconsistent framing can invite both inventorship objections and sufficiency challenges.
How do the EPO and Italian courts assess inventive step and sufficiency? The EPO applies the problem-solution approach, asking whether the claimed solution makes a non-obvious technical contribution; AI involvement does not lower this bar. Sufficiency requires disclosure enabling the skilled person to reproduce the invention, which is harder where the result depends on an opaque model or proprietary training data. Claims drafted around a self-enabling technical result are more robust.
Because AI disputes turn so heavily on how a result was produced, evidence preservation often determines the outcome before a court is ever involved. Both rights-holders and defendants need disciplined protocols, and the two sides have different but complementary obligations.
A rights-holder who anticipates asserting or defending an AI-related patent should preserve the full technical record that establishes how the invention was made and when. Practical steps include:
A defendant must balance the duty to preserve relevant evidence against the risk of over-production. A defensible document-retention policy, applied consistently and suspended by a litigation hold once a dispute is reasonably anticipated, protects against allegations of evidence destruction while avoiding the disclosure of more than is necessary. Defendants should also be alert to the burden of proof: in many disputes it is the claimant who must establish infringement and entitlement, so a defendant’s preservation strategy can focus on the evidence that undermines those elements rather than on reconstructing the claimant’s entire development history.
Enforcing AI patents italy requires choosing the right forum and sequencing national and EPO proceedings to maximum effect. The available routes are complementary, and sophisticated strategy coordinates them rather than treating them in isolation. Holders of European patents should also consider the Unified Patent Court (UPC), which, for patents not opted out, provides an additional, centralised venue for infringement and revocation across participating EU states including Italy.
Patent infringement in Italy is litigated before the specialised business court divisions (sezioni specializzate in materia di impresa). A rights-holder can seek urgent interlocutory relief, including preliminary injunctions to stop infringing activity pending trial, and measures for the description and seizure (descrizione and sequestro) of infringing goods and evidence. These interim tools are particularly valuable in AI disputes, where the ability to secure a defendant’s technical materials early, before they are altered or deleted, can be decisive. Italian procedure also permits the appointment of court-appointed technical experts (consulenti tecnici d’ufficio), whose role in assessing complex AI and software questions is central to the outcome.
Marshalling persuasive expert evidence, and preparing the technical narrative so that a court-appointed expert can follow it, is a core part of enforcement strategy.
Where the goal is to challenge the validity of a granted European patent, the EPO opposition procedure offers a centralised route that can remove or narrow a patent across all designated states. Opposition must be filed within nine months of the mention of grant; the subsequent appeal lies to the Boards of Appeal. These are the forums in which patentability arguments, lack of inventive step, insufficiency, added matter, are tested against EPO practice. For AI-related patents, an opposition focused on insufficiency and the plausibility of the claimed technical effect can be especially effective. Coordinating an EPO opposition with national litigation requires care, because the outcomes and timing of the two tracks interact.
For clients with manufacturing operations or supply chains spanning several jurisdictions, enforcement of AI patents italy is rarely confined to Italy alone. Coordinating injunctive relief across jurisdictions, deploying customs measures to intercept infringing goods at the border, and sequencing actions to obtain the fastest effective relief all form part of a cross-border strategy. The interaction between national interim measures, the UPC and centralised EPO challenges should be planned from the outset so that a win in one forum is not undercut by a parallel proceeding in another.
What steps should rights-holders take to enforce or defend AI-related patents in Italy? Preserve the technical evidence early, model snapshots, logs and datasets with a documented chain of custody. Instruct technical experts to prepare replicability and prior-art analysis. Consider urgent interlocutory relief, including preliminary injunctions and description or seizure measures, before the specialised Italian business courts. Where validity is in issue, weigh an EPO opposition, and coordinate national, UPC and cross-border measures to secure effective relief.
A defendant or opponent facing an AI-related patent has a familiar but potent set of invalidity grounds, each of which acquires a particular flavour in the AI context. The strongest arguments typically combine lack of inventive step, insufficiency of disclosure, added matter and, where relevant, excluded subject-matter such as a claim directed to a computer program or mathematical method as such without a technical contribution.
Expert evidence is the engine of an AI invalidity case. Replicability testing can demonstrate that the claimed result cannot be reproduced from the disclosure alone, supporting an insufficiency attack. Analysis of training data can reveal that the AI output was the predictable product of known inputs, undermining inventive step, and can surface prior art embedded in the data itself. Challenges to algorithmic transparency can expose whether the patent truly enables the invention or relies on an undisclosed model. Assembling credible, independent experts who can perform and explain this testing is central to a successful challenge to AI patents italy.
Beyond substantive invalidity, defendants should assess procedural defences, questions of entitlement where inventorship was poorly documented, or added-matter objections arising from amendments during prosecution. In many cases the commercial realities favour a negotiated outcome: a licence, a cross-licence, or a narrowing of the claims. A clear-eyed assessment of the strength of the invalidity case, informed by early expert input, allows a defendant to choose between fighting to revoke the patent and settling on favourable terms.
Most disputes over AI patents italy can be avoided, or substantially de-risked, by robust contracting before any invention is made. Agreements that touch AI-assisted R&D should address IP ownership and assignment in express terms, grant licences with clearly defined scope, and provide for model and algorithm escrow where continuity or reproducibility matters. Audit rights allow a party to verify how an AI tool was used and what data informed it. Warranties and indemnities should allocate the risk of third-party IP claims arising from training data or model outputs.
For procurement and RFP teams, a standard checklist that requires suppliers to confirm assignment of deliverables, disclose material use of AI, and accept indemnity obligations will close the most common gaps before they become litigation.
To translate this guidance into action on AI patents italy, counsel and R&D leaders should work through a short, concrete programme:
AI patents italy will remain a fast-moving field through 2026 and beyond, but the practical fundamentals are already clear enough to act on. The inventor must be a human; ownership of AI inventions is decided by contract and statute; patentability still turns on a demonstrated technical contribution; and disputes are won or lost on preserved evidence and credible expert analysis. Organisations that audit their agreements, document the human role in AI-assisted invention, and build litigation readiness before a dispute arises will be far better placed than those that wait for the law to settle.
The time to strengthen your position on AI patents italy is now, while the underlying rules remain stable and the strategic choices are still yours to make.
This article was produced by Global Law Experts. For specialist advice on this topic, contact Francesco Misuraca at SMAF & Associates, SAS, S.T.A., a member of the Global Law Experts network.
posted 51 seconds ago
posted 22 minutes ago
posted 42 minutes ago
posted 1 hour ago
posted 2 hours ago
posted 2 hours ago
posted 3 hours ago
posted 3 hours ago
posted 3 hours ago
posted 3 hours ago
posted 4 hours ago
posted 4 hours ago
Find the right Legal Expert for your business
Send welcome message