With the development of patent applications on artificial intelligence (AI) comes a number of legal issues. These include whether the criteria for patentability are fulfilled when it comes to AI-based inventions, the treatment of innovations induced by an AI mechanism that is itself patented (invention of inventions), and the liability regime for the potentially negative effects of such innovations. This paper first provides an overview of these issues. In a second stage, we re-examine the economic foundations of the patent mechanism in order to propose elements of answers. At the end of this reflection, it appears that a strict interpretation of the claims of a patent related to an AI mechanism is likely to be compatible with the legal issues.