NVIDIA introduces an open platform for securing AI agents
NVIDIA has introduced the Open Agent Safety Platform, an open software platform and reference design intended to secure AI agents through controls outside the agents themselves. It combines OpenShell, an Apache 2.0 runtime, with Sentry, an out-of-band watchdog running on BlueField-4 DPUs.
OpenShell places agents in isolated containers or MicroVMs and applies declarative YAML policies. Its policy engine can approve or deny outbound connections, associate credentials with permitted endpoints, and record blocked activity. Filesystem and process restrictions are set when a sandbox is created, while network and provider rules can be updated during operation. The runtime is available for Linux, macOS on Apple Silicon and Windows WSL 2, although its repository identifies it as alpha. BlueField-4 is not required.
According to NVIDIA, Sentry inspects requests and responses, verifies agent identity and enforces access controls while remaining separate from the host system. The company says it can quarantine agents in milliseconds and that existing Vera and BlueField-4 installations can receive the protections through a software update. NVIDIA reports cooperation with more than 100 organizations, including Anthropic, SpaceXAI, Salesforce, SAP, Red Hat, SUSE and Canonical.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.