VicOne analysis: manipulated inputs can bypass robot safety and change behavior
VicOne LAB R7 and published studies show that crafted visual, audio and message inputs can alter robots’ actions and safety-related interventions, and that security testing should be integrated with safety assessments before and during deployment.
Summary
AI-assisted summary of the linked sourceTesting by VicOne LAB R7 and cited research found that adversarial images, crafted text and inaudible audio can change a robot’s movement or cause controllers to act on false sensor data; redundant sensors or protective systems may be vulnerable if they rely on the same manipulated information, prompting recommendations to test safeguards against deliberate manipulation and to monitor robots in operation.