Consent in Context: Why One-Time Consent Falls Short in Adaptive Camera Framing

The growing use of intelligent camera systems in video communication, where systems autonomously zoom, pan, and reframe to include participants, raises privacy concerns. These AI-powered systems can capture bystanders, mirror reflections, or others who have not provided their consent. This dynamic behavior exposes the limitations of static one-time consent models. In this work, we empirically demonstrate how adaptive framing mechanisms can be spoofed to include unintended individuals, and we present findings from a user study with 50 participants. Participants reported that static consent felt insufficient and expressed a strong preference for real-time notifications or control mechanisms when the framing changed. Our results underscore the need to rethink consent as a dynamic, continuous process, especially for AI systems that alter their behavior based on environmental sensing. We argue that privacy-by-design in adaptive camera systems must account for fluid participation boundaries and ensure that users retain control over what the camera captures during operation.