The emerging global governance of artificial intelligence (AI) is shaped by numerous political actors. Inviting non-state actors into such processes is typically assumed to address a perceived democratic deficit, by promoting increased representation, transparency, and openness. In the AI sphere, however, non-state actors include the same multinational companies that develop the technology to be regulated. Surprisingly, the task of normatively theorizing the democratic role of non-state actors in global AI governance has nevertheless been largely ignored. This paper addresses this by specifying, first, under what conditions non-state actors contribute to the actual democratization of global AI governance, as ‘democratic agents’, and second, under what conditions they instead contribute to the strengthening of the prerequisites for future democratization, as ‘agents of democracy’. We conclude that, although few non-state actors are authorized to act as ‘democratic agents’, their exercise of ‘moral’, ‘epistemic’, and ‘market authority’ could make them legitimate ‘agents of democracy’.