Are AI models’ weights protected databases?

The ongoing AI revolution is based on machine learning models that rely on vast artificial neural networks trained on enormous amounts of data. These models use a transformer architecture and are generated through an automated process of training. The model is comprised of two files, a simple run file that establishes the model architecture, and a much larger file of parameters or weights expressed as floating point numbers. Protecting copyright in the machine learning model weights faces a challenge due to their functional nature. However, the sui generis protection for databases established in Directive 96/9 provides a strong candidate for protecting model weights. To apply the EU sui generis database protection to machine learning models’ weights, one needs to establish that these can qualify as a database and that there is substantial investment in them by the database maker.

Source: Kluwer Copyright blog