French criminal law lacks a definition of AI and French substantive law does not provide for AI having legal personality or capacity. French legal literature denies to this day the possibility of an independent accountability of AI. The analysis of classical figures of criminal Law (actus reus, mens rea and possible forms of punishment for AI) shows that the AI cannot be seen per se as a perpretator of an offence but rather as an instrument for committing an offence. The central role of human beings must therefore be emphasized: criminal liability due to AI could accordingly be considered in the context of a specific mechanism of attribution to natural or legal persons in relation to AI. This leads to the reflexion on the theoretical basis for accountability of AI. Specific laws regarding vehicles with delegated control regulated for the first time the interaction between man and AI in French substantive law and provided a first practical example. The field of compliance is also of great interest in organising how to anticipate the criminal risks related to AI activity through the classical tools of compliance (risk-mapping, code of conduct, detailing the various safety and transparency obligations for designers/manufacturers, regulation through dedicated authorities).