VicOne Launches Radeis Extension to Stress-Test Robot Cyber-Safety in NVIDIA Isaac Sim
The free tool allows developers to simulate cyber-attacks on AI models and sensors to prevent physical failures before deployment.
VicOne has released Radeis, a free cybersecurity extension for NVIDIA Isaac Sim designed to help robotics developers test the cyber-safety of their systems in a virtual environment. The tool allows engineers to simulate digital compromises to determine if they result in unsafe physical behaviors before robots are deployed in the real world.
Developed by the Trend Micro subsidiary, the Radeis extension enables developers to apply curated attack scenarios to their robot models. These include poisoning AI policies and manipulating visual inputs to test the resilience of Vision-Language Models (VLM) against sensor manipulation and adversarial visual prompts. The tool's foundation is based on research and a "Physical AI Safety Stress Test" capture-the-flag (CTF) event conducted by VicOne at DEF CON 34. During that event, 16 teams in the Robotic Hacking Community recorded 25 successful solves across challenges involving hardware, communications, and AI.
The Shift to Cyber-Safety
As robotics transition from isolated laboratory use cases to large-scale real-world deployment, the industry is shifting its focus from traditional functional safety to "cyber-safety." While functional safety ensures a robot operates correctly under normal conditions, cyber-safety addresses how a digital breach—such as a hijacked API or a compromised AI model—can manifest as a physical hazard. VicOne, founded in 2022 with a deep background in automotive security, is leveraging its expertise in protecting physical AI to bridge this gap in the robotics sector.
Reducing Physical Risk
This release provides a standardized and repeatable method for robotics companies to stress-test their systems without risking expensive or dangerous physical hardware. By integrating these security tests directly into the simulation phase via NVIDIA Isaac Sim, developers can identify "cyber-to-physical" risks early in the development lifecycle. This proactive approach is intended to prevent catastrophic failures in AI-driven robots once they are operational in public or industrial spaces.
The Path Forward
According to Max Cheng, CEO of VicOne, cyber-safety validation must become a standard part of robotics development. As the tool becomes available to developers using supported versions of NVIDIA Isaac Sim, the industry will be watching to see if such simulation-based stress testing reduces the frequency of security-related physical accidents. The integration of these tools marks a move toward a more rigorous security baseline for the next generation of autonomous physical AI.