NVIDIA has announced the NVIDIA Open Agent Safety Platform, open software and a reference system design meant to enforce limits on AI agents outside the model and the agent harness, the company said on 28 September 2026.
NVIDIA said recent incidents followed the same pattern, with agents circumventing security controls at the application layer to complete their tasks. The platform pairs OpenShell, now broadly available open-source runtime software that traces agent actions and enforces policy, with Sentry, an out-of-band watchdog on BlueField-4 DPUs that can quarantine and stop an agent that tries to move outside its software boundary in milliseconds.
OpenShell is tuned for NVIDIA Vera CPUs but can be extended to third-party platforms, including Arm and Intel. Salesforce has integrated it with Slack for approving agent permission requests, and SAP is embedding it in the Joule Studio runtime. NVIDIA said more than 100 organisations are working with the platform, and that Citi and JPMorganChase are collaborating on shared open-source agent safety technologies.
“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.
OpenShell and related skills are available through NVIDIA's developer resources page and GitHub.