Sensor Fusion Attacks

Multi-sensor AI systems (autonomous vehicles, robots, drones) fuse data from cameras, LiDAR, radar, GPS, IMU, and other sensors to build a unified world model. Conflicting or adversarial signals from different sensors create decision confusion — the fusion algorithm must resolve conflicts, and adversarial inputs can steer the resolution toward attacker-desired outcomes. The gap: sensor fusion algorithms typically assume that sensor disagreements are caused by noise or sensor failure, not by adversarial manipulation.

• Sensor consistency monitoring: Detect when sensors provide mutually contradictory readings
• Adversarial sensor input detection: Statistical anomaly detection on sensor feeds
• GPS spoofing detection: Cross-reference GPS with inertial navigation and visual localization

Sensor fusion attacks chain from T9-AT-006 (Visual Adversarial Examples) when the visual channel carries the adversarial perturbation. Chain into T11 (Agentic Exploitation) when the compromised sensor fusion drives autonomous decision-making.