Teams using AI agents should account for the personal inferences agents can make and consider how those inferences may spread across connected systems.