Microsoft has just released as an open-source project an automation tool for security testing AI/ML systems [here the announcement].

Bruce Schneier has published an analysis of a world where AI/ML systems can be hackers [here the paper].