VicOne shows off Physical AI cybersecurity demos at NVIDIA GTC Berlin

VicOne unveiled two robotics cybersecurity demonstrations at NVIDIA GTC Berlin on Oct. 23, 2026, showing how teams can test behavior-changing cyber risks before deployment and respond to threats while robots are operating. The demos tie cybersecurity more directly to robot safety as AI-driven systems move from simulation into real-world action.

Why it matters: - VicOne is pitching cybersecurity as a safety layer for physical AI, where manipulated inputs can push robots into unsafe behavior. - The demonstrations show how robotics teams can evaluate whether safeguards still hold when an attacker changes the information a robot uses to decide and act. - The timing matters for European robotics teams preparing for the EU Machinery Regulation, which applies from Jan. 20, 2027.

What happened: - VicOne, a Japan-headquartered cybersecurity company focused on physical AI, presented two demonstrations at NVIDIA GTC Berlin on Oct. 23, 2026. - One demo showed how robotics teams can detect and respond to runtime threats that could alter robot behavior during operation. - The second demo showed how teams can test cyber manipulation effects in simulation before deployment. - VicOne said the demonstrations combined NVIDIA technologies with VicOne’s Physical AI cybersecurity tools.

The details: - The runtime demo used NVIDIA Nemotron models to support contextual reasoning inside VicOne Rthena. - Rthena assessed runtime risk and coordinated approved responses across a robot fleet. - In the scenario, a robot transporting medication near a pediatric ward encountered suspicious commands and anomalous communication signals. - Rthena monitored runtime activity to detect malicious intent and assess potential behavioral risk. - When it detected a threat, Rthena could recommend or apply adaptive protections such as real-time prompt patching and policy enforcement. - VicOne said those actions were meant to neutralize malicious instructions before they affected the robot’s decisions or actions. - The simulation demo used the VicOne Radeis Extension with NVIDIA Isaac Sim to test how adversarial visual input could change robot behavior. - In the demo, manipulated visual input redirected a robot’s attention, causing it to ignore a stop sign and keep moving. - The Radeis Extension let teams study the behavior change in simulation and gather evidence before physical testing and release. - VicOne’s free Radeis Extension for NVIDIA Isaac Sim is available now. - The broader Radeis and Rthena platforms are commercially available. - VicOne said its tools help teams assess cyber risks and build evidence relevant to a robot’s design and operating conditions. - The company also said the approach can help maintain safe, reliable behavior across real-world deployments. - VicOne’s about page says the company has uncovered 180+ zero-days across automotive and robotics, adds 100M+ new threat intelligence signals monthly, and has 30 U.S. patent applications pending in AIDV and AI security. - VicOne says it is headquartered in Tokyo and is a Trend Micro subsidiary. - More information is available on VicOne’s website. - VicOne also shared social links to its LinkedIn page and YouTube channel.

Between the lines: - The demos frame robot cybersecurity as a pre-deployment and in-operation safety test, not just a defensive IT function. - That matters because robotics failures can become physical hazards, so proving resilience against manipulation becomes part of safety evidence. - VicOne is also positioning its tools around regulatory preparedness, especially for systems sold in Europe. - The use of simulation suggests robotics teams may be able to reduce risk before exposing hardware to real-world environments.

What's next: - Robotics teams can use the free Radeis Extension now for NVIDIA Isaac Sim testing. - Companies using the broader Radeis and Rthena platforms can apply the tools to runtime monitoring and operational response. - European robot makers will face the EU Machinery Regulation starting Jan. 20, 2027, making safety-relevant cyber protections more relevant in design and validation workflows.