The Missing Layer in Robot Safety Assurance

Robot safety assurance has traditionally focused on preventing failures when systems malfunction, but a new concern has emerged: robots can be manipulated through cyberattacks that change what they perceive, decide, or do while appearing to function normally. These attacks can occur across three layers of a robot's system: corrupting AI models with hidden triggers during training, exploiting vulnerabilities in the underlying system infrastructure to override commands, and manipulating sensor inputs or reasoning at runtime. Addressing this risk requires extending safety assurance beyond functional safeguards to include cybersecurity practices throughout a robot's lifecycle, from design through operation, so that robots remain within safe boundaries even when under deliberate attack.