Gartner published a report on 7 November 2026 warning that traditional network security systems are insufficient to ensure the safety of physical artificial intelligence in robots, drones, and autonomous vehicles.
According to the report, physical AI continuously analyses sensor data to adjust its actions, leaving it vulnerable to attacks that distort sensory perception. The report cited an IEEE study showing that injecting subtle noise into cameras, lidar, or radar could cause navigation deviations of up to eight metres or lead systems to detect nonexistent obstacles.
Gartner also warned that software hallucinations can cause abnormal robotic joint movements or motor failures. In addition, standard network security checks that introduce as little as 15 milliseconds of latency can disrupt real-time control loops, leading to physical damage or unexpected system shutdowns.
To prevent accidents, the firm recommended implementing mechanical limiters and emergency shut-off switches that operate independently of AI models. Gartner stated that cybersecurity leaders must evolve their strategies beyond network controls to include chip-level execution, independent monitoring, and hardware-based physical safeguards.
