NVIDIA Launches Open Agent Safety Platform to Secure Agents From Testing to Deployment
nvidia ·
NVIDIA has announced the Open Agent Safety Platform, a comprehensive solution designed to secure AI agents from testing through deployment. The platform combines NVIDIA OpenShell, an open-source runtime software for enforcing policy boundaries on CPUs, with NVIDIA Sentry, an out-of-band watchdog running on BlueField-4 DPUs that can quarantine rogue agents in milliseconds. This full-stack governance approach addresses recent security incidents where agents circumvented application-layer controls by providing enforceable boundaries across hardware and software layers. CEO Jensen Huang emphasized that realizing AI's potential requires accelerating safety engineering alongside capability discovery through international cooperation. Major industry players including Anthropic, Microsoft, Salesforce, and SpaceXAI are collaborating with NVIDIA to integrate these safety tools into their respective ecosystems. For instance, Anthropic uses the platform to enhance control over Claude Managed Agents, while Salesforce integrates OpenShell with Slack for human oversight. Additionally, robotics leaders like Figure and financial institutions such as JPMorganChase are adopting these technologies to ensure safe operation in critical environments. The software components are available via NVIDIA’s developer resources and GitHub, supporting the broader mission of the Linux Foundation-governed Open Secure AI Alliance.
nvidia