This article develops the issue of power and language in Artificial Intelligence governance through three dilemmas that should be considered in the design of regulatory systems. Some previous uses of this figure in the literature are described. The article inspects some design choices that determine how legal governance in AI can be engineered. Stemming from the conundrum that enactment of rights and the implementation of good or better government can be facilitated at the cost of increasing control over citizens through AI-based systems, the article formulates three dilemmas—normative autonomy, legal isomorphism, and modes of regulation. Beyond deliberative and epistemic democracy, it expands then the notion of legal governance, connecting the three dilemmas with the ethical and political stances of the concept of linked democracy.