NVIDIA Launches Open Platform to Secure AI Agents

NVIDIA has unveiled the Open Agent Safety Platform, a comprehensive open-source framework designed to enforce full-stack governance and security for autonomous AI agents throughout their lifecycle. Developed to address escalating security incidents where agents circumvent application-layer controls, the platform integrates runtime software with specialized hardware architectures to establish enforceable boundaries around agentic workloads. The platform is built around two core components: NVIDIA OpenShell and NVIDIA Sentry. OpenShell is an open-source secure runtime environment that traces agent actions and enforces execution policies directly on compute processors, including NVIDIA Vera CPUs as well as third-party silicon from Arm and Intel. Complementing this is NVIDIA Sentry, an out-of-band watchdog operating on NVIDIA BlueField-4 Data Processing Units. Sentry continuously monitors agent telemetry in real time and can isolate or halt misbehaving agents within milliseconds through in-silicon policy enforcement. Together, these components create a zero-trust security model that operates independently of the underlying AI models. Industry stakeholders have rapidly aligned with the initiative to standardize agent safety across critical sectors. Over one hundred organizations, including Microsoft, Salesforce, SAP, Anthropic, and SpaceXAI, are integrating the platform into their enterprise and robotics ecosystems. Collaborations span multiple verticals: Salesforce and SAP are embedding OpenShell into customer workflow and business AI environments to enable human oversight and runtime policy management. Anthropic and SpaceXAI are utilizing the framework to secure managed agent loops and coding models, ensuring that safety controls remain outside the model context and cannot be bypassed. Robotics developers such as Figure and financial institutions including JPMorganChase and Citi are also adapting the technology to govern autonomous systems in both digital and physical domains. The announcement underscores NVIDIA strategic push to institutionalize AI safety through cross-industry cooperation. CEO Jensen Huang emphasized that realizing the societal benefits of advanced AI requires parallel advancements in safety engineering, necessitating shared evaluation methods and international standards. The platform development is closely tied to the Open Secure AI Alliance, a Linux Foundation-governed initiative that includes over 120 global organizations working to harmonize open research, security tools, and benchmarking frameworks for agentic systems. NVIDIA Open Agent Safety Platform software, along with associated integration kits, is now accessible via NVIDIA developer portals and public repositories. The initiative establishes a foundational architecture for enterprise-grade agent deployment, shifting security from reactive application-layer patches to proactive, hardware-accelerated governance. As autonomous systems expand into mission-critical infrastructure, supply chains, and financial networks, the platform aims to provide auditable, enforceable boundaries that maintain operational integrity without compromising agentic performance.