On 10 June 2026, the Italian Council of Ministers approved two draft legislative decrees aimed at adapting national legislation to Regulation (EU) 2024/1689 (the AI Act). Among the proposed measures is an amendment to Article 98 of the Italian Industrial Property Code governing the protection of trade secrets.
The Content of the New Paragraph 1-bis
The proposal expressly includes data, algorithms and mathematical methods used for training AI systems among the business information and technical-industrial know-how that may qualify as trade secrets, provided that the traditional requirements of secrecy, economic value and reasonable protection measures are met.
Scope of the Provision: Inclusion and Definition
The amendment operates on two complementary levels. First, it expressly clarifies that AI-related assets may fall within the scope of trade secret protection. Secondly, it provides a definition of algorithms and mathematical methods, including model architectures, optimisation functions, training procedures and configurations, as well as any other technical-computational element functional to the development of AI systems.
A Clarification Rather than a New Intellectual Property Right
The amendment does not create a new intellectual property right over AI models, nor does it alter the existing requirements for trade secret protection. Rather, it provides legal certainty by confirming that AI assets may benefit from the existing trade secret regime.
Objective Limits of Protection
Protection remains conditional upon the fulfilment of the requirements set out in Article 98. Consequently, publicly available model weights, openly accessible datasets, and generally known architectures or optimisation functions will not automatically qualify for protection.
Why Trade Secrets Rather than Copyright?
The legislative choice reflects the difficulties of relying on copyright law to protect AI assets. Copyright remains tied to human authorship and individual creative contribution, whereas many AI-related assets derive their value primarily from investment, data collection, engineering effort and technical development rather than creative expression.
From this perspective, trade secret protection appears better suited to preserving the economic value of large-scale AI investments through an objective and technology-neutral legal framework.
Procedural Aspects: The Specialised Business Courts
Disputes concerning training datasets, algorithms and model weights benefiting from trade secret protection would fall within the jurisdiction of the Italian Specialised Business Courts, ensuring continuity with existing case law on trade secrets and confidential business information.
Critical Perspectives: The Transparency Challenge
The proposal is not immune from criticism. Some commentators argue that recognising AI data, architectures and parameters as potential trade secrets may increase opacity at a time when both EU and national legislation are moving towards greater transparency and explainability requirements.
However, a distinction must be drawn between the explanation of a specific automated decision and disclosure of the underlying model. The former concerns the intelligibility of a decision affecting an individual, whereas the latter concerns access to datasets, model weights, architectures and training configurations. These are distinct legal and technical issues.
A more delicate tension may arise where verification of an AI system requires access to training data or the model itself in order to investigate bias, discrimination or malfunction. In such circumstances, trade secret protection may operate not as a barrier to explanation, but as a limitation on full inspection of the system.
Conclusion
By expressly recognising the potential trade secret protection of AI assets, the proposed amendment provides businesses with valuable guidance for structuring research and development investments and designing protection strategies. It confirms a clear policy choice in favour of an objective form of protection based on secrecy rather than a creativity-based model centred on authorship. The effectiveness of this choice will ultimately depend on how courts and regulators balance innovation, legal certainty, transparency and accountability in the AI ecosystem.
For more information, please visit our Intellectual Property and Digital and Data practice areas.
