ABSTRACT
Deep neural networks are a dominant artificial intelligence technology targeted for regulation in both the EU and other jurisdictions. The correct legal qualification of this technology as an object of legal relations becomes a necessary condition for the proper application of the new rules, as well as for properly solving the issues of protection of rights or determining the content of contractual obligations. The article analyses the features of artificial neural networks and assesses whether these objects can be considered copyrighted objects, whether information about these objects can be recognized as a trade secret and whether the weights of trained artificial neural network models can be stored as databases. The results of the study reveal that the artificial neural network programming code, when only API accesses are used, will not meet the copyright protection criteria, and the hyperparameters of the networks may constitute a protected commercial secret if they are not made public and not obvious to specialists in the relevant field. Meanwhile, the weights of trained artificial neural networks can be protected as databases because they meet the criteria of separability, individual accessibility and autonomous value, but such protection does not apply in all jurisdictions.
Stasys Drazdauskas, Artificial neural network as an object of legal relationships (2025) 15 Queen Mary Journal of Intellectual Property (4) 400-419. Published: 1 December 2025.
Leave a Reply