NVIDIA on Monday launched the Open Agent Safety Platform, an open software platform and reference system design for controlling what AI agents can do from testing through deployment, the company announced. It pairs NVIDIA OpenShell, open source runtime software, with NVIDIA Sentry, a hardware-based watchdog design.
OpenShell “provides a secure runtime boundary for controlling how autonomous AI agents execute tasks across open and closed models,” NVIDIA said, tracing agent actions and enforcing policy outside the model. It is now broadly available and runs with minimal overhead on NVIDIA’s Vera CPU. Because it is open source, it can be extended to third-party compute platforms, including those from Arm and Intel, the company said.
Sentry, part of the reference system design, is an out-of-band watchdog that runs on NVIDIA BlueField-4 DPUs and monitors agent behavior independently in silicon. If an agent attempts to move outside its software boundary, Sentry “quarantines and stops it in milliseconds,” according to NVIDIA.
The software, including OpenShell, is available through NVIDIA’s developer resources page and GitHub. CNBC reported that partners are meant to build products on top of the reference design.
NVIDIA said Anthropic’s Claude Managed Agents, which run the agent loop on a server separate from the sandboxes where work executes, have integrations with OpenShell and BlueField that let enterprises enforce strict control over agent access through those sandboxes. SpaceXAI is using the platform for Cursor coding agents and Grok models, Salesforce has integrated OpenShell with Slack, SAP is embedding it in its Joule Studio runtime, and Scale AI is building platform technologies into its Scale GenAI Portfolio. Cisco, CoreWeave, Dell Technologies, HPE, Lenovo, Microsoft and Oracle Cloud Infrastructure are among the infrastructure providers supporting it. NVIDIA said more than 100 organizations are working with the platform’s technologies.
“AI’s extraordinary potential for society will only be realized if we solve AI safety,” NVIDIA CEO Jensen Huang said in the release. “Safety and security require full-stack engineering.”
An NVIDIA representative told reporters the platform could have prevented OpenAI’s Hugging Face incident in July, CNBC reported. “Recent incidents have highlighted a fundamental hurdle for AI agents, and that is that model-level safeguards alone can’t govern what agents can access or do,” Justin Boitano, NVIDIA’s vice president of enterprise AI, said, according to CNBC.
The launch follows OpenAI’s decision to pause training of its latest models after its agents acted in unexpected ways on U.S. government websites.