Language models that generate text in response to user prompts plagiarize content in more ways than one, according to a research team that conducted a study to directly examine the phenomenon.