Image Movement Attacks on Optical See-Through HMDs: Covert Gaze Manipulation and Privacy Risks in AR/MR Systems

Optical see-through AR/MR head-mounted displays (HMDs) are increasingly adopted for enhanced task efficiency and user experience, but also introduce new security and privacy threats. This paper investigates "Image Movement Attacks," a form of Visual Perceptual Manipulation Attack (VPMA) where visual elements on an HMD are subtly repositioned to covertly influence gaze and head orientation, and to potentially cause privacy intrusion or user awareness manipulation. This paper makes three key contributions: it considers realistic attack scenarios, provides experimental validation, and explores fundamental countermeasures. As one of the realistic attack scenarios, an attack on the supply chain is presented, such as embedding malicious programs into assets used for application development. A user study with 22 participants using the Magic Leap 1 device showed that continuous visual movements often went unnoticed, posing significant risks for covert gaze manipulation. Our results indicate that covert image movements alter gaze and head orientation, raising concerns that Image Movement Attacks in AR/MR HMDs could trigger serious issues in an impactful attack chain, or manipulate even everyday decisions. To mitigate these risks, we explore a certification framework for third-party assets, as well as platform-level anomaly detection, using monitoring and integrity checks.