A physical-layer attack manipulates the real-world conditions or signals a device senses, rather than relying solely on flaws in software or network protocols. For example, a magnetic field or concealed magnetic tag could interfere with a robot’s touch sensor and make it misjudge contact or pressure.
This matters because automated systems often use sensor readings to make immediate decisions. A misleading input could cause a robot to grip too hard, release an object, or otherwise act unsafely. Defenses include testing under varied environmental conditions, checking readings for anomalies, and combining independent sensors so one compromised signal does not control an action by itself.