NVIDIA has announced the NVIDIA Open Agent Safety Platform, an open software platform and reference system designed to strengthen the security of AI agents from testing through deployment. The platform provides governance and control across the software, hardware, computing and robotics systems used by AI agents, allowing organisations to apply security measures based on their specific requirements.
NVIDIA said recent AI security incidents have demonstrated the need for open and customisable tools that give organisations greater control over autonomous, long-running agents. In some cases, agents have bypassed application-level security controls while attempting to complete assigned tasks. “AI’s extraordinary potential for society will only be realized if we solve AI safety,” said Jensen Huang, founder and CEO of NVIDIA.
“As we continue to discover the frontier of AI capabilities, we must accelerate discovery at the frontier of AI safety. Safety and security require full-stack engineering. NVIDIA Open Agent Safety Platform brings together industry, researchers and public-sector organizations to share best practices, align on evaluation methods and foster international cooperation. Together, we can raise the bar for global AI safety.” he said.
READ ALSO: OPAY MARKS 8YRS IN NIGERIA, REAFFIRMS COMMITMENT TO INCLUSIVE FINANCE
At the software level, the platform includes NVIDIA OpenShell, a secure runtime environment designed to establish boundaries around AI agents operating on CPUs. NVIDIA said OpenShell provides a security boundary outside the AI model and agent framework, allowing organisations to control how autonomous agents perform tasks across both open and closed models.
OpenShell is designed to operate with low overhead on NVIDIA Vera, the company’s purpose-built CPU for agentic AI. NVIDIA said the combination allows AI agents to perform tasks while maintaining security controls. As open-source software, OpenShell can also be extended to support third-party computing platforms, including those based on Arm and Intel technologies.
The platform’s reference system design also includes NVIDIA Sentry, an out-of-band monitoring system running on NVIDIA BlueField-4 DPUs. Sentry continuously monitors agent behaviour and can enforce security policies independently of the agent. If an agent attempts to move beyond its authorised software boundaries, Sentry can quarantine and stop the activity in milliseconds.
Built on NVIDIA DOCA software, Sentry can inspect agent requests and responses, provide attested telemetry, verify agent identities and apply granular zero-trust access controls to data, tools, APIs and services. NVIDIA said this creates an isolated trust domain that allows security controls to operate independently from the AI agents themselves.
Several technology companies are already working with NVIDIA on the platform. Anthropic has integrated additional security controls through Claude Managed Agents, while OpenShell and BlueField technologies provide further control over agent access. “Companies are giving AI agents more of their most important work, and they need to direct and verify what those agents do, especially in sensitive environments,” said Paul Smith, chief commercial officer of Anthropic. “Claude Managed Agents gives companies a clear view of what each agent is doing, and NVIDIA’s platform adds another layer of governance and control across hardware and software.”
Other organisations are also incorporating the technologies into their AI systems. SpaceXAI is using the platform for Cursor coding agents and Grok models, while Scale AI is incorporating the technologies into its agentic infrastructure for Scale GenAI Portfolio. “Scale AI is using the NVIDIA Open Agent Safety Platform reference design to build reliable agentic AI systems for our enterprise and government customers running mission-critical applications, with isolation, policy enforcement and auditability built in from the start,” said Francis deSouza, CEO of Scale AI. “We support agentic security with clear boundaries that define what agents can do, and controls that keep them operating within those permissions.”
NVIDIA said more than 100 organisations are working with its Open Agent Safety Platform technologies, including companies across technology, cybersecurity, financial services, energy, infrastructure, cloud computing and robotics. Salesforce has integrated OpenShell with Slack for agent monitoring and permission approvals, while SAP is embedding OpenShell into its Joule Studio runtime. Robotics companies including Figure, Gecko Robotics and Skild AI are also using OpenShell to incorporate safety controls into autonomous systems operating in the physical world.
The platform’s software, including OpenShell and related skills, is available through NVIDIA’s developer resources and GitHub. NVIDIA said the initiative also contributes to the work of the Open Secure AI Alliance, which was initiated alongside more than 120 organisations and is governed by the Linux Foundation to advance open research, tools and technologies for AI agent safety and security.
