Inside NVIDIA Sentry: Secure AI Agents at Runtime | NVIDIA Technical Blog
Oct 21, 2026Ofir Arkin

Summary
This post details the technical implementation of NVIDIA Sentry, a security framework for AI agents running on BlueField-4 DPUs. It introduces DOCA Tower as a transparent proxy leveraging Envoy for TLS termination and Layer-7 inspection, integrated with Open Policy Agent (OPA) for policy evaluation in under a millisecond. DOCA Argus is presented as a host-independent memory introspection tool for runtime context gathering, mapping containers, processes, and their properties. The system correlates network traffic with runtime context (e.g., container hashes, process integrity, SBOM) to enforce granular access policies, with in-silicon enforcement for rapid threat containment.
More from the Secure AI Agent Deployment trail
- How NVIDIA OpenShell Reference Architecture Secures and Governs AI Agent Workloads at Scale | NVIDIA Technical Blog
- NVIDIA Open Agent Safety Platform: A Reference for Continuous In-Silicon Agent Monitoring | NVIDIA Technical Blog
- Add Runtime Controls to AI Agents with NVIDIA OpenShell | NVIDIA Technical Blog
- Building an Adaptive Agentic Cybersecurity System with NVIDIA Nemotron | NVIDIA Technical Blog
- Where Security Fits in an AI Agent Stack | NVIDIA Technical Blog