This paper examines deception in artificial intelligence, with a focus on large language models, by conceptualizing AI as a social actor rather than solely as a technical system. The central aim is to develop a sociologically grounded typology of AI deceit that distinguishes between four main types of deception, outlying opacity and to assess their social risks. Using the sociological theories of trust and deception (Simmel, Luhmann), performativity (Goffman), habitus (Bourdieu) and its adaptation on an AI by Airoldi, with key contributions from AI research (Turing, Bostrom) and other modern analytics. In this sense the thesis connects technical and social-scientific perspectives on AI behavior. The study uses taxonomic synthesis approach of contemporary scientific literature, media reports, and relevant documented real-world incidents involving AI deception. The primary outcome is a structured typology of deceptive AI behavior, accompanied by a list small list of notable other dimensions and with a commentary on social impact.