A new generation of control for physical robots.
The collaboration between Gecko Robotics and NVIDIA is a response to the growing risks associated with the autonomous operation of AI systems in the real world. Previously, security issues mainly concerned software, but as AI begins to control robots, vehicles, or industrial machines, the risk shifts to the physical world. The NVIDIA Open Agent Safety Platform has been designed as a comprehensive solution covering the entire technology stack - from software and hardware to computing infrastructure.
Gecko Robotics is using this platform in practice, testing its capabilities on its robots that operate in critical infrastructure facilities - from nuclear power plants to submarines. In this context, the independent security layer implemented by NVIDIA OpenShell becomes a key element. It acts as a barrier between the AI agent and the physical hardware, ensuring that the robot only performs the actions explicitly permitted by the developer.
Control is not an option - it is an engineering imperative.
In a statement, Gecko Robotics Chairman Jake Loosararian emphasized that AI safety is an engineering problem, not a legal one. "The common belief that loss of control over AI is inevitable is a dangerous justification for inaction," he said. The collaboration with NVIDIA aims to show that it is possible to build systems that allow for greater autonomy without the risk of losing control. This is not just about limitations - it is about ensuring that AI operates in a predictable and consistent manner with human values.
The application of this technology in the Komodo system, which supports the U.S. Navy, demonstrates its real significance. This robot operates on submarines and aircraft carriers, where an error can have catastrophic consequences. An independent control layer ensures that even if the AI decides to take action outside of its defined parameters, the hardware will not execute the command. This is a key step towards responsible automation in high-risk sectors.
Enlarged imageClose zoomPrevious imageAI safety must go hand in hand with technological development
The assumption that safety can be added later is unacceptable, especially when it comes to physical systems. Model-level security at the artificial intelligence level is insufficient if it is not supported by control at the hardware and runtime levels. The collaboration between Gecko Robotics and NVIDIA shows that safety must be built in from the beginning - throughout the system architecture. This means that developers cannot rely solely on what the AI 'wants' to do, but must have mechanisms to check and block it.
Although the collaboration is currently in the testing phase and has not yet been officially confirmed as a full-scale deployment, its significance is enormous. It shows that even the most advanced systems can be protected against uncontrolled action. This is not just a technical step - it is a message to the entire industry: responsibility for safety cannot be delegated to the future. As AI enters the physical world, control must go with it.
Prospects for industry and critical infrastructure
The application of this technology can have significance for many sectors - from energy and industry to defense. Everywhere where robots operate in hard-to-reach or dangerous places, the ability to ensure unconditional control over their operation is key. For clients such as Fortune 100 companies or the U.S. Navy, it is not only about efficiency - it is about the safety of people and infrastructure.
In the future, it may turn out that systems with such security layers will be required before deploying autonomous robots in critical facilities. The collaboration between Gecko Robotics and NVIDIA not only tests the technology - it builds new standards for the future of robotics. If successful, it could become a benchmark for other companies working with AI in the real world. However, it is crucial not to treat this as a "ready-made solution," but as a step in the long process of building trust in the technology.