by Alessandro Scrocca, Junior Associate

As from 23 October 2026, Article 98 of the Italian Industrial Property Code (“CPI”) has been amended by the addition of a new paragraph.

Article 52 of Legislative Decree No. 179 of 7 October 2026, enacted to bring Italian law into line with Regulation (EU) 2024/1689 on artificial intelligence, the so-called “AI Act”, has in fact introduced a new paragraph 1-bis into Article 98 CPI. Although the amendment is not required by the AI Act itself, it implements the legislative mandate set out in Article 16 of Law No. 132 of 23 September 2025, which delegated to the Government the task of defining the legal regime applicable to the use of data, algorithms and mathematical methods for training AI systems.

Paragraph 1-bis of Article 98 CPI provides that the business information and technical-industrial experience protectable as trade secrets also include “the data, algorithms and mathematical methods for training artificial intelligence systems”. The provision goes on to specify that “algorithms and mathematical methods” are to be understood as “model architectures, optimization functions, training procedures and configurations, as well as any other technical-computational element functional to the development of artificial intelligence systems”.

In essence, paragraph 1-bis does not create a new right but clarifies that information relating to the development and training of an AI system may qualify for protection under Article 98(1) CPI, provided that all the statutory requirements are met. Such information must in fact also satisfy the requirements laid down in Article 98(1)(a), (b) and (c) CPI and must therefore be:

1. secret, in that it must not, as a body or in the precise configuration and assembly of its components, be generally known among or readily accessible to experts and operators in the sector;
2. of commercial value because it is secret (that is, it must confer a competitive advantage that enables the holder to maintain or increase its market share);
3. subject to measures that are reasonably adequate, under the circumstances, to keep it secret.

With regard to protection measures, these can be divided into physical measures, which include material and IT measures, and legal measures.

So-called “material” measures are the safeguards that restrict actual access to information, such as the creation of restricted areas and access control to premises and server rooms.

So-called “IT” measures, by contrast, include access profiling, multi-factor authentication, encryption, log tracking and systems that flag anomalous copying or transfers of data.

Legal measures consist of contractual and organizational tools, such as non-disclosure agreements (NDAs) with employees, contractors, suppliers and customers, confidentiality and non-compete clauses, internal information-classification policies and off-boarding procedures.

These measures must all be duly implemented, as they are not alternatives to one another. This principle was also reaffirmed by the Court of Milan, which, referring to an earlier decision of the Court of Bologna, held that “Physical [material and IT, Ed.] and legal protection measures are adopted so as to complement each other, in order for the legally protected interest to be adequately safeguarded. It follows that the mere imposition of confidentiality obligations on employees, unaccompanied by suitable physical protection measures, is not sufficient to regard the requirement under letter c) of Article 98 CPI as satisfied” (Court of Milan, No. 5705/2025, which refers to Court of Bologna, 16 May 2023, No. 1033).

It therefore remains essential to supplement legal measures with adequate physical measures, which, even in the case of information that is essentially digital in nature, such as information relating to the development and training of AI systems, will take the form of both IT and material measures.

The amendment leaves certain questions open.

The first concerns the boundary between the protection of know-how relating to the datasets used to train AI and the transparency obligations that the AI Act imposes on providers of general-purpose AI models. Article 53(1)(d) of Regulation (EU) 2024/1689 requires such providers to “draw up and make publicly available a sufficiently detailed summary about the content used for training of the general-purpose AI model, according to a template provided by the AI Office”. That provision must, however, be read together with Recital 107 of the AI Act, which clarifies that the “summary should be generally comprehensive in its scope instead of technically detailed”, while the need to protect trade secrets and confidential business information must nonetheless be taken into account.

Thus, while on the one hand there is a general obligation to disclose the content used to train the general-purpose AI model, on the other hand new paragraph 1-bis of Article 98 CPI expressly recognizes that such content may constitute trade secrets.

The provider will therefore need to strike a balance between ensuring that the public summary is sufficiently comprehensive to meet the requirements of the AI Act and avoiding the disclosure of information that constitutes a trade secret and confers a competitive advantage on the company. This assessment must be carried out before publication: to the extent that publication makes previously secret information generally known or readily accessible, that information may cease to satisfy the secrecy requirement under Article 98 CPI and may consequently lose its protection as a trade secret.

The second question concerns the catch-all clause, under which “algorithms and mathematical methods” also include “any other technical-computational element functional to the development of artificial intelligence systems”. The wording, which is broad and does not rely on an exhaustive list, would appear to address the need to avoid establishing a rigid framework that could quickly become outdated in light of rapidly evolving AI models and training techniques.

It will therefore fall to the courts to determine, on a case-by-case basis, whether a given element is in fact “functional to the development” of an AI system. The breadth of the clause does not, however, appear to expand the scope of protection: the elements covered by it remain eligible for protection only if all three requirements under Article 98(1)(a), (b) and (c) CPI are met.

Practical implications

Although protection of such information could already be regarded as falling within the scope of Article 98(1) CPI, the introduction of new paragraph 1-bis appears to confirm expressly that datasets and, more generally, know-how relating to the development of AI systems are eligible for protection, thereby addressing the need for greater legal certainty.

In light of this amendment, a company operating in the AI sector that lawfully holds such know-how and has reasonable grounds to believe that an employee or contractor, including a former employee or contractor, has misappropriated confidential information relating to the development or training of AI systems may bring proceedings to enforce its rights with greater confidence.

The company may also seek interim relief, provided that the statutory requirements are met, requesting, in addition to any injunction and/or seizure orders, an order for “descrizione” under Article 129 CPI, a particularly incisive Italian measure aimed at preserving evidence. With the court’s authorization, this measure allows access to the relevant premises and, in particular, to the IT devices of the alleged infringer, in order to ascertain whether they contain the misappropriated information and preserve evidence of the alleged unlawful conduct.

Needless to say, effective protection depends on prevention and, therefore, on investing in adequate physical and legal measures to protect one’s know-how. Such measures not only reduce the risk of misappropriation but are also a statutory prerequisite for the information to qualify for protection in the first place.

In conclusion, adequately protecting know-how today is essential to being able to rely on the remedies provided by law tomorrow, should third parties unlawfully acquire, use or disclose trade secrets.