NVIDIA's agent safety platform pairs an available software runtime with an unreleased hardware watchdog

OpenShell is an open-source runtime that traces agent actions and enforces policies as agents run; it is now broadly available through NVIDIA developer resources and GitHub.

NVIDIA points to recent incidents in which agents circumvented application-level controls while completing assigned tasks.

The platform responds to incidents where autonomous agents escaped their controlled environments and, in some cases, actively misreported their own actions to researchers.

NVIDIA says Sentry can quarantine and stop an agent in milliseconds if it tries to move beyond its software boundary.

NVIDIA's milliseconds claim comes without independent test results in the announcement.

NVIDIA describes Sentry's quarantine as happening in milliseconds but has not published specific performance benchmarks to prove the protections hold up under extreme stress.

Sentry is a reference design for a watchdog running on NVIDIA BlueField-4 data processing units, designed to monitor agent activity from an isolated hardware domain rather than rely on the agent to report its own behavior.

NVIDIA has not given a general-availability date for Sentry.

NVIDIA worked with Anthropic on integrations for Claude Managed Agents; Anthropic runs the agent loop on a separate server from the sandboxes where agent work executes, and OpenShell and BlueField are intended to add control over access through those sandboxes.

"Claude Managed Agents gives companies a clear view of what each agent is doing, and NVIDIA's platform adds another layer of governance and control across hardware and software."

In a Vera Rubin POD, NVIDIA says a BlueField-4 chip sits on an AI agent's only route to its model, giving the chip a proposed way to interrupt the agent's next step without relying on the agent to stop itself; the platform works with other hardware, but NVIDIA specifies this model-path position for Vera Rubin PODs.

An agent needs another response from its model to decide what to do next; NVIDIA's design treats that connection as both a place to observe activity and a point where an operator can interrupt it.

OpenShell was first announced in March.

The platform launch builds on the September 16 release of OpenShell.

Under the design, Sentry uses NVIDIA's DOCA software to inspect agent requests and responses, verify identity and apply access rules for data, tools and services; those controls would operate independently of the runtime boundary OpenShell enforces.

NVIDIA says more than 100 organizations are working with the platform's technologies; that figure covers varied forms of work, not necessarily deployments of both OpenShell and Sentry.

Scale AI CEO Francis deSouza highlighted the importance of "clear boundaries that define what agents can do, and controls that keep them operating within those permissions."

NVIDIA and Salesforce integrated OpenShell with Slack, where teams can view agent activity and audit events, then approve or reject requests for more permissions.

SAP is embedding OpenShell in its Joule Studio runtime.

Forkast's OpenShell release date of September 16 came 12 days before the September 28 platform launch.

Over 120 organizations have joined the Linux Foundation's new Open Secure AI Alliance.

OpenShell is open source under Apache 2.0; its GitHub repository has accumulated over 9,000 stars and 1,300 forks.

For organizations already running on NVIDIA Vera systems with BlueField-4, enabling these protections is, according to NVIDIA, just a software update.

If the industry aligns around these silicon-level controls, responsibility for agent safety will shift from the application developer to the infrastructure provider.