The development of ICT has created an information environment in which vast amounts of data flow through digital space, algorithms shape exposure to content, and an increasing share of material is generated by artificial intelligence, raising new risks for critical infrastructure, public safety, and democratic processes. Riding this wave, the gatekeeping practice of mainstream media imperceptibly shifts toward gatewatching, at times contributing to the spread of crises. Technically, the recognition and detection of AI-generated “fake news” has become a hot topic, with startups and major technology companies investing heavily in solutions. This paper examines AI-generated “fake news”, used here as a shorthand for synthetic disinformation and related deceptive content, that function as inductors of crises. Drawing on the BANI (Brittle, Anxious, Nonlinear, Incomprehensible) perspective, the study analyzes a corpus of 217 research articles on AI-generated disinformation and semi-structured interviews with state officials and decision-makers to (i) identify crisis-inducing types of AI-generated “fake news”, (ii) compare them with an established fake-news typology, and (iii) propose an updated framework tailored to AI-generated content and crisis contexts. Findings indicate six crisis-inducing types: (1) deepfakes; (2) false statements or notorious misinformation; (3) content with a fictitious source; (4) favorable reports or statements; (5) conspiracy theories; and (6) satire without clear labeling. The paper outlines implications for detection pipelines, critical-infrastructure protection, and journalistic practice, and suggests directions for future research on linking AI-generated “fake news” to specific crisis categories, escalation dynamics, and early-warning and response playbooks.