How open-source tech can be quietly manipulated to mess up apps

French outfit Mithril Security has managed to poison a large language model (LLM) and have made it available to developers to prove a point about misinformation.…