Abstract Lexical ambiguity is a constitutive property of natural languages and plays a central role in human communication. In French, a large number of lexemes can be associated with several meanings and present several distinct senses, the interpretation of which depends on the linguistic, discursive, and pragmatic context. While human speakers generally resolve these ambiguities intuitively, their processing by artificial intelligence systems remains a major challenge. This article offers an in-depth study of lexical ambiguity in French, combining a theoretical linguistic approach with a critical analysis of contemporary language models. Through the examination of concrete cases, we analyze the mechanisms of lexical disambiguation, the performance of AI systems, and the inherent limitations of distributional approaches to meaning.