Security researchers, including VicOne LAB R7, show that robots can be steered by corrupted inputs. In tests, text on a poster was read as an instruction and changed a robot dog's path. Crafted inaudible audio changed a simulated hospital robot's behavior. In another case, an injected message moved a robot that should have stayed still.
Robots already share hallways, hospitals, and factory floors with people. Their safety features — slowing down, stopping, keeping balance — only work if the data feeding them is honest. If a false reading or faked message can blind those safeguards, the physical risk is real. This reframes robot safety as a cybersecurity problem, not just an engineering one.
✦ A robot is only as safe as the truthfulness of its inputs.