2026-11-05 - Algorithms and training data as trade secrets: the amendment to the Industrial Property Code in Legislative Decree 179/2026
This contribution closes the series block devoted to
Legislative Decree no. 179 of 7 October 2026
, the second decree implementing Italian Law 132/2025. Chapter IV of the decree, Articles 52 and 53, addresses a topic that has remained in the background of public debate but has direct consequences for anyone developing AI systems: the protection of data, algorithms and training methods as trade secrets.
What it provides
Article 52 of the decree inserts a paragraph 1-bis into Article 98 of the
(in Italian). Data, algorithms, and mathematical methods for training AI systems are included among protected business information and technical-industrial experience, provided the conditions set out in paragraph 1 are met: secrecy, commercial value derived from secrecy, and reasonable steps to keep it secret.
The new paragraph specifies that, for training purposes, algorithms and mathematical methods mean model architectures, optimization functions, training procedures and configurations, and any other technical-computational element functional to the development of the systems. Disputes remain within the jurisdiction of the specialized business divisions of the courts, under Article 134(1)(b) of the Code (Article 53).
Press release no. 185 of the Council of Ministers
(in Italian) did not mention these provisions, which the published text confirms in the draft’s wording.
Conditional protection
The provision does not create a new right: it clarifies that this information can fall within the existing protection, under the same conditions as any other confidential business information and consistently with
on trade secrets. The decisive point, therefore, remains the third condition: anyone seeking protection must show that they have taken adequate steps to keep the information secret. For AI providers, this is an organizational matter before it is a legal one: access controls, contracts, information classification, traceability.
The choice also serves a systemic function. Mathematical methods are not patentable under Article 45 of the Industrial Property Code; trade secret protection covers elements that patents do not.
The interaction with access to evidence
The most interesting point is the link with
. As I explained in the
contribution devoted to civil liability
, in actions for damage caused by AI systems, the court may order disclosure of logs, technical documentation and information on human oversight, taking trade secrets into account and, where there is a risk of disclosure, adopting protective measures under Article 121-ter of the Industrial Property Code.
The express inclusion of data, algorithms and mathematical methods for training AI systems among the information that may constitute a trade secret makes clear that those measures can be invoked for them too. It does not mean removing the information from the disclosure order: Legislative Decree 160/2026 asks the court to strike a balance, not to exclude. But the defendant will have a stronger argument for asking that access be limited to a small number of people or that documents be made available in redacted form.
The interaction with the AI Act
Classification as a trade secret does not exempt anyone from the obligations of
. Technical documentation, deployer information, and summaries of the content used to train general-purpose AI models remain due. The AI Act also requires authorities and bodies receiving information to respect its confidentiality, including trade secrets, under Article 78. Trade secret protection therefore does not neutralize the AI Act’s obligations: it affects confidentiality safeguards and arrangements for access to information, while it applies fully to third parties and, in litigation, to the balance struck with the disclosure order.
Concluding remarks
The provision strengthens AI developers’ position and meets a genuine need for legal certainty. Its value will depend, however, on businesses’ ability to document the confidentiality measures adopted, and its impact will be measured above all in litigation, where it will have to coexist with the push towards transparency that Legislative Decree 160/2026 has introduced.
For those wishing to navigate the regulatory framework referred to in this contribution, the
of this site are available: the
, the
and the
.
The reference sources are available on the institutional websites:
Legislative Decree no. 179 of 7 October 2026
, the
draft sent to the Chamber of Deputies
(in Italian), the
on Normattiva (in Italian),
,
and
.
New to Gemini? This capsule is best read with a Gemini client, such as Lagrange.