NVIDIA Sentry
A trusted foundation for AI agents.
Overview
What Is NVIDIA Sentry?
NVIDIA Sentry is a unified security software framework built on NVIDIA DOCA™ and running on NVIDIA BlueField™-4 DPUs to protect AI agents at runtime. Operating from a hardware-isolated, host-independent security domain, NVIDIA Sentry provides continuous agent governance, out-of-band observability into agent behavior and reasoning, runtime threat detection, and in-silicon policy enforcement. NVIDIA Sentry can detect deviations from defined security policies and behavioral profiles and contain threats within milliseconds, independently of the agent and host software. NVIDIA Sentry integrates with NVIDIA OpenShell™ and enterprise cybersecurity solutions to provide layered agent security from software to silicon.
Benefits
The Trust Layer for AI Agents
Powered by DOCA, NVIDIA Sentry runs on BlueField to deliver out-of-band security that keeps AI agents governed, observable, and contained at runtime.
Trusted Security for Confident Agent Deployment
Running on BlueField-4's hardware-isolated, host-independent security domain, NVIDIA Sentry maintains security visibility and control even if an agent escapes its software boundaries.
Continuous Agent Governance, Observability, and Protection
NVIDIA Sentry combines agent identity and delegated authority with visibility into agent behavior, reasoning, and interactions to detect deviations and enforce defined security policies.
Security at AI Agent Speed
NVIDIA Sentry uses BlueField-4's in-silicon security capabilities to continuously evaluate agent activity. It enables containment and quarantine within milliseconds when defined boundaries are crossed.
Efficient and Scalable Agent Security
BlueField-4 provides dedicated infrastructure processing for NVIDIA Sentry, preserving host compute resources and reducing the CPU tax to increase agent capacity across AI deployments.
Developer-Ready and Extensible Design
Powered by DOCA libraries and microservices, NVIDIA Sentry enables developers, cybersecurity teams, and security providers to integrate their own signals, policies, and detection logic.
Ecosystem-Backed Security Solution
Leading security providers integrate NVIDIA Sentry's trusted agent telemetry, detection, and enforcement capabilities into their solutions, giving enterprises a proven foundation for securing agentic AI.
安全な AI エージェントのための NVIDIA OpenShell リファレンス アーキテクチャ
NVIDIA OpenShell 安全なランタイム、Vera CPU、Vera Rubin システム、および BlueField-4 上で動作する NVIDIA Sentry を基盤とするフルスタックのオープン アーキテクチャ。
Technology
Explore NVIDIA Sentry
Built on BlueField and powered by DOCA, NVIDIA Sentry delivers out-of-band, in-silicon security for AI agents at runtime.
Builders
Building the World’s Secure Agent Infrastructure
Leading enterprise cybersecurity providers are using NVIDIA Sentry to deliver out-of-band agent governance, deep observability, threat detection, and in-silicon enforcement at the scale their customers need.
Next Steps
Ready to Get Started?
Build AI agent security with NVIDIA DOCA, the software foundation for NVIDIA Sentry.
Access DOCA Documentation
Explore the complete DOCA documentation library, including SDK guides, API references, and release notes.
Download DOCA
Download the NVIDIA DOCA SDK and access APIs, libraries, and sample applications to start building on BlueField.
FAQs
NVIDIA Sentry is a unified security software framework built on NVIDIA DOCA that runs on NVIDIA BlueField to protect AI agents at runtime. Operating from a hardware-isolated, host-independent security domain, NVIDIA Sentry provides continuous agent governance, out-of-band observability, runtime threat detection, and in-silicon policy enforcement, independently of the agent and host software it protects.
NVIDIA Sentry continuously monitors agent behavior and interactions across models, data, tools, networks, and services. It combines agent identity and delegated authority with runtime security signals to determine whether agents are operating within their defined boundaries. When NVIDIA Sentry detects anomalous behavior or policy violations, it can enforce policy in real time and quarantine agents within milliseconds, while providing trusted telemetry to enterprise security platforms for investigation and response.
Unlike host-based security solutions, NVIDIA Sentry operates from BlueField-4's hardware-isolated, host-independent security domain, outside the agent and host software it protects. NVIDIA Sentry doesn’t rely on the agent or host to report its own behavior, maintaining visibility and control even if an agent escapes its software boundaries.
NVIDIA Sentry continuously evaluates agent activity and behavioral signals and uses BlueField's in-silicon security capabilities to enforce policy directly in the infrastructure data path. When defined boundaries are crossed, NVIDIA Sentry can restrict agent access or isolate communications within milliseconds, without waiting for the agent or host software to take action.
Yes. Leading enterprise cybersecurity providers are integrating NVIDIA Sentry's trusted agent telemetry, detection, and enforcement capabilities into their solutions. Powered by DOCA libraries and microservices, NVIDIA Sentry enables security teams and providers to integrate their own signals, policies, and detection logic, giving enterprises a proven foundation through the security platforms they already use.
NVIDIA Sentry is built on NVIDIA DOCA and runs on BlueField—on the node’s only path to the model—providing continuous, out-of-band observability into agent behavior and enforcing security policies in real time at line speed. NVIDIA Sentry is part of the NVIDIA Open Agent Safety Platform built on the NVIDIA OpenShell Reference Architecture for Safe AI Agents.