Of the countless challenges posed by artificial intelligence to criminal law, the identification of human liability for harms ‘caused’ by machines is among the most difficult. The use of ‘positions of guarantee’ in their traditional form is particularly problematic, mainly due to the difficulty of ascertaining the causal link between a specific human error and the harmful result produced by the artificial agent. The report aims to strike a fair balance between the effectiveness of the system and the need to respect fundamental criminal law principles. Where it is impossible to identify the liability of the human agent, the liability of the legal entity remains conceivable.