Machine learning systems, now often referred to as artificial intelligence or AI systems, have been instrumental in shaping the public sphere for at least two decades. The advent of search engines, and, a few years later, algorithmically curated social media sites, has created a new information ecosystem with substantially different properties than the legacymedia system based mainly on newspapers, radio, and television. Not all of these characteristics are desirable from a normative point of view. For example: Ample research shows that mis- and disinformation have a strategic advantage compared to factually correct information in social media environments. The advent of generative machine learning systems that can create realistic text, images, and increasingly also audio and video content will exacerbate this problem: the marginal cost for creating convincing disinformation at scale is dropping rapidly. This article cites some examples that have already been observed and discusses possible mediation strategies.