{
  "id": 16771635,
  "title": "Your Robot’s Safety Functions Already Work. What If the Input Lies?",
  "url": "https://urgent.news/2026/10/27/your-robots-safety-functions-already-work-what-if-the-input-lies",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-10-27T16:14:39.000Z",
  "source": {
    "name": "IEEE Spectrum",
    "slug": "ieee-spectrum",
    "url": "https://spectrum.ieee.org/ai-robot-safety-cybersecurity-vicone"
  },
  "original_language": "en",
  "account": "When robots navigate their environment, they rely on various inputs, such as distance readings, position estimates, and stop signals. However, if one of these inputs is incorrect or manipulated, the robot may still respond as initially programmed. For instance, a mobile robot might interpret a hallway as clear even when someone is present. Similarly, a collaborative arm may estimate that a worker is farther away, and a humanoid robot might fail to recognize a stop signal when a person crosses its path. These discrepancies could arise from faulty inputs or deliberate manipulation. Researchers at VicOne LAB R7 investigate how such manipulated inputs can alter robot behavior and impact safety systems. A robot processes more than just raw data; it interprets visual and auditory information to determine its actions. Consequently, any information placed in the robot's environment can influence its behavior. Research has demonstrated this phenomenon in vision-language-action (VLA) models, where subtle alterations in camera input caused simulated robots to fail tasks. Furthermore, adversarial images could cause the models to disregard subsequent instructions. In a practical test, a robot dog treated text on a poster as an instruction, changing its movement. Similarly, manipulating audio in a hospital-service-robot simulation changed the robot's behavior without altering its assigned tasks. These examples highlight how manipulated inputs can transform unintended outcomes into repeatable attacks. Safety-related decisions for robots depend on various inputs, including readings, estimates, and messages. If any of these inputs are manipulated, the robot's protective systems may still intervene, but the intervention could be based on incorrect information. For instance, injecting a ROS 2/DDS message into a robot under a safe-control setting caused it to move, demonstrating the potential vulnerability of robotic systems. Similarly, sound at a sensor's resonant frequency could distort a humanoid robot's reported rotation, leading to incorrect balance adjustments. Researchers have demonstrated acoustic attacks against drones with vulnerable gyroscopes, though similar attacks causing humanoids to fall have not been confirmed. To ensure robot safety, security teams must evaluate whether existing protections can detect malicious inputs independently of manipulated information. Deliberate attacks challenge traditional safety assumptions, as attackers can introduce false inputs at their discretion. This flexibility necessitates reassessing safety evaluations to account for intentional manipulation. Additionally, redundant sensors require thorough testing, as a single manipulated input could affect multiple safety checks simultaneously. In autonomous driving, a crafted physical object misled a system combining camera and LiDAR perception, highlighting the need for comprehensive safety testing. Cyber threats introduce intent into the safety equation, requiring updated risk assessments that consider attack timing and repetition. Safety teams must evaluate the evidence behind speed, protective distance, and operating area limits, including how manipulated information could affect these limits. Standards such as China's GB/T 45502-2025, IEC TS 63074, and the EU Machinery Regulation provide guidance on securing safety-relevant systems and data. VicOne LAB R7 offers a comprehensive approach to testing and monitoring robot safety across the lifecycle. Through the Robotic Hacking Community, VicOne helps teams investigate cyber threats and validate potential effects before deployment, ensuring that safety systems remain robust against manipulation.",
  "summary": "This article is brought to you by VicOne . A mobile robot slows for someone in a hallway. A collaborative arm eases its movement as a worker approaches. A humanoid pauses to let a person pass. Each response depends on information about the robot or its surroundings: a distance reading, a position estimate, or a stop signal. But what happens when one of those inputs is wrong, and the robot still…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}