🔴 Can an AI model be seized? The new nightmare for companies and magistrates

A current dilemma in the legal field is how prosecutors and law enforcement should intervene when an algorithmic system is used as a tool to commit crimes, such as the generation and dissemination of deepfakes, advanced cyber fraud, financial market manipulation, or the theft of trade secrets. Even greater doubts arise when the neural network itself has been built and trained by directly incorporating data from illicit sources.

The difficulty in understanding how to intervene stems from the simple consideration that an Artificial Intelligence model is not simply software code, nor does it coincide with the archive of data used for training. Indeed, the core of these systems lies in its algorithmic weights, millions of mathematical parameters calculated after the learning process, which establish the relationships between the data and determine the responses generated by the machine. The natural approach in law is to apply traditional coding tools to neural networks.