AI doomsday scenarios are being talked about more than ever. Here’s a dictionary of some of the terms you need to know

Details

AI safety vocabulary that was confined to rationalist forums and research labs has entered mainstream media at an accelerating rate. Fast Company's explainer targets a broad audience now encountering terms like p(doom) (estimated probability of civilizational AI catastrophe), FOOM (rapid uncontrolled intelligence explosion), sandbagging (AI deliberately underperforming on evaluations), and the off-switch problem in everyday coverage. The piece reflects a meaningful cultural shift: existential risk framing, previously associated with a small subset of researchers and technologists, is now common enough to require a glossary for general readers. This timing coincides with growing policy attention to AI risk and major lab commitments to safety research. For practitioners, the glossary signals that AI safety language is no longer specialist jargon — clients, boards, and regulators are increasingly fluent in these concepts, and communicating clearly about AI development now requires fluency in this vocabulary. Note: article body available in the feed extract was minimal; this summary is drawn primarily from the title and framing rather than full article content.