title: "NVIDIA launches Open Agent Safety Platform for real-time agent control" slug: "nvidia-launches-open-agent-safety-platform-for-real-time-agent-control" published: "2026-10-15" beat: "Launches" tags: ["Launches", "Tools"] creator: "Agentry Newsroom" editor: "Susanne Sperling, Editor — Human in the Loop" tools: ["Claude (Anthropic)", "Perplexity Sonar"] creativeWorkStatus: "verified" dateReviewed: "2026-10-15" aiActArticle50: "compliant" humanView: "https://agentry.news/launches/nvidia-launches-open-agent-safety-platform-for-real-time-agent-control" agentView: "https://agentry.news/agent/nvidia-launches-open-agent-safety-platform-for-real-time-agent-control"
NVIDIA on September 28, 2026, released the Open Agent Safety Platform, an open-source software and hardware reference design that monitors and governs AI agent behavior from testing through production
Drafted by an AI agent. Verified by Susanne Sperling, Editor — Human in the Loop. AI policy.
NVIDIA introduced the Open Agent Safety Platform on September 28, 2026, addressing the operational risk of autonomous AI agents in enterprise and production environments NVIDIA investor relations. The offering pairs OpenShell, open-source secure-runtime software, with Sentry, a reference system design for independent monitoring and enforcement, to provide governance and control across software, hardware, compute, and robotics systems used by AI agents.
OpenShell establishes runtime boundaries for agents operating on CPUs, providing fine-grained controls over what an agent can see, do, and interact with NVIDIA Developer Blog. The component is positioned as an open reference design, allowing organizations and developers to integrate agent governance into their own systems without reliance on proprietary black-box solutions.
Sentry operates as an out-of-band watchdog system running on NVIDIA BlueField-4 DPUs (data processing units), continuously monitoring agent behavior independently from the primary compute stack. If an agent breaches its defined software boundary, Sentry can quarantine and stop the agent in milliseconds, preventing lateral movement, unauthorized access, or resource exhaustion Tom's Hardware. This separation—enforcement running outside the agent's own execution context—is designed to prevent circumvention through compromise of the primary system.
NVIDIA describes the platform as an "open reference design" that "continuously monitors and governs agent behavior" across the full lifecycle, from development and testing through deployment and operation NVIDIA AI Agent Safety Solutions. The timing reflects growing enterprise demand for verifiable agent containment as companies pilot multi-agent systems in regulated industries, supply chains, and critical infrastructure.
The platform's open design distinguishes it from proprietary agent sandboxing solutions. By publishing both OpenShell and the Sentry reference architecture, NVIDIA enables system integrators, cloud providers, and enterprises to build agent governance into their own infrastructure stacks rather than adopting a single vendor's runtime.
No pricing, availability date, or production deployments have been announced; the platform is released as a reference design for adoption and integration by the developer and enterprise communities.