NVIDIA unveils new safety platform amid rising security concerns towards rogues AI agents

NVIDIA has launched the NVIDIA Open Agent Safety Platform, an open software initiative designed to secure AI agents from testing through deployment with comprehensive governance across hardware, compute, and robotics systems. The announcement comes as recent security incidents have highlighted vulnerabilities where AI agents circumvent application-layer security controls to complete assigned tasks. Jensen Huang, founder and CEO of NVIDIA, emphasized that “AI's extraordinary potential for society will only be realized if we solve AI safety”, positioning the platform as an effort to bring together industry, researchers, and public-sector organizations to align on evaluation methods and foster international cooperation on global AI safety standards.

The platform comprises two core components: NVIDIA OpenShell, an open source software providing a secure runtime boundary that traces all agent actions on NVIDIA Vera CPUs, and Sentry, a reference system design running on NVIDIA BlueField-4 DPUs that acts as an out-of-band watchdog monitoring agent behavior in real time. Sentry operates independently in silicon, capable of quarantining agents attempting to breach their boundaries within milliseconds, an approach described as “in-silicon security enforcement” that remains invisible to both agents and potential attackers. The open nature of OpenShell allows it to be extended for compatibility with third-party compute platforms from competitors including ARM and Intel.

Industry partners have already begun integrating the platform into their operations. SpaceXAI is deploying NVIDIA Open Agent Safety Platform for Cursor coding agents and Grok models, with company president Mike Nicolls noting that “safety should be enforced outside the model by additional controls the agent can't get past”. Anthropic's chief commercial officer Paul Smith also endorsed the initiative, highlighting that while Claude Managed Agents provides companies visibility into agent activities, NVIDIA's platform adds “another layer of governance and control across hardware and software”. The collaborative approach reflects growing recognition among AI developers that securing increasingly autonomous systems requires full-stack engineering solutions rather than relying solely on model-level safeguards.