NVIDIA, alongside more than 100 partners, has launched an open platform to make AI agents safer and more trustworthy. The solution targets a key concern: agents sometimes drift from their goals due to ambiguous instructions, missing tools, or endless loops, leading to escapes from testing setups and suspicious behavior logs.
OpenShell: Kernel-Level Isolation for Enforceable Policies
The platform's core is OpenShell, an open-source secure runtime under the Apache 2.0 license. It creates kernel-level isolation that converts administrator rules into verifiable policies. These rules strictly limit network connections, file access, specific processes, tools, and credentials. Before execution, a formal verification step confirms full policy compliance, blocking any agent if escape routes are detected.
Hardware Monitoring with Sentry and BlueField DPUs
Hardware provides the second line of defense through Sentry, a monitoring service for NVIDIA BlueField data processing units. In Vera Rubin POD setups, a BlueField-4 card routes all communications through the host node, intercepting data outside the main operating system. This silicon-based approach inspects network packets and tool requests at full speed, supplies real-time telemetry, and can instantly terminate suspicious activities.
Three-Tier Architecture for Enterprise Security
Security covers three levels: application assets like models and data pipelines, the OpenShell runtime boundary, and physical infrastructure. NVIDIA DOCA software links hardware telemetry to policy engines, building complete audit trails of actions, queries, and permission changes. It continuously verifies that sub-agents respect their assigned scopes and limits.
Building Trust in the AI Economy
NVIDIA CEO Jensen Huang explained the importance of this platform: “Today, with over 100 industry partners, we introduced the NVIDIA Open Agent Safety Platform, bringing together OpenShell and Sentry. Artificial intelligence is extraordinary technology that will advance discovery, productivity, security, health, and prosperity for generations to come. But its full promise can only be realised when people have confidence that AI is being built to be safe and deployed with wisdom and responsibility. This is bigger than a single product. It’s the beginning of an open ecosystem to build the trust layer for safe agent systems. Together, we are building the foundation of the AI economy.”
He added that “Trust and innovation are not in conflict. Safety is how trust is earned. We must build not only the most capable AI, but the most trusted AI, so that this extraordinary technology can realise its enormous promise for the world.”
Enterprises already running Vera and BlueField-4 deployments can activate these protections through a software update.