The article proposes an approach to building probabilistic estimates of the spread of a social engineering attack between two users, which takes into account the informational influence of users. The theoretical significance of the study lies in creating a basis for the subsequent modeling of multi-step social engineering attacks. The practical significance lies in the formation of a base for software automation of identification of the most probable and critical scenarios for the spread of social engineering attacks, and, as a consequence, the creation of a tool that helps decision-makers to take effective measures to eliminate vulnerabilities.