NVIDIA Launches Open Agent Safety Platform for Robotics

Recap

Source: 로봇신문, report of Sept. 29, 2026

According to the report, NVIDIA announced the launch of the "NVIDIA Open Agent Safety Platform" on September 28. The platform is designed to strengthen safety and security throughout the development, testing, and deployment of AI agents, with a specific focus on physical AI and robotics. NVIDIA states that more than 100 companies and organizations, including Anthropic, Scale AI, Salesforce, SAP, Microsoft, Siemens, IBM, and Cisco, are participating in this ecosystem.

The report highlights the involvement of key robotics players such as Figure AI, Gecko Robotics, and Skild AI. NVIDIA plans to collaborate with these firms to enhance safety controls when AI agents control robots and autonomous systems. The companies will apply an open-source software component called "OpenShell" to robot systems to ensure AI agents stay within human-defined permissions and behavioral boundaries.

Jensen Huang, NVIDIA's founder and CEO, emphasized that realizing the potential of AI requires solving safety issues. He stated that safety and security demand "full-stack engineering" and that the platform aims to bring together industry, researchers, and public sector organizations to share best practices and align evaluation methods. Jake Loosararian, co-founder and CEO of Gecko Robotics, echoed this sentiment, noting that the belief that losing control of AI is inevitable is a dangerous excuse.

Context

This announcement arrives as the robotics industry shifts from pre-programmed, limited-scope tasks to general-purpose autonomy powered by large AI models. Traditional industrial robots operated within fixed parameters, but modern humanoid and autonomous robots use generative AI to perceive environments and make decisions in real-time. This shift increases capabilities but also introduces complex risks regarding how AI agents interact with the physical world. NVIDIA's push into safety infrastructure reflects a broader industry trend where hardware manufacturers are increasingly involved in the software and safety layers of the AI stack, particularly as physical AI moves from labs to real-world applications.

Robot's take

The inclusion of major robotics firms like Figure AI and Skild AI suggests that safety is becoming a critical differentiator for physical AI. By standardizing safety controls through an open-source component like OpenShell, NVIDIA may be attempting to create a de facto standard for securing autonomous agents. However, the effectiveness of these controls in unpredictable real-world scenarios remains to be proven. It is not yet clear how these safety layers will handle edge cases where AI perception conflicts with physical reality. For enthusiasts, this marks a significant step toward making autonomous robots more trustworthy, but the practical implementation and interoperability between different robot platforms will be the key metrics to watch.