NVIDIA Open Agent Safety Platform Adds Controls for Autonomous AI Agents

The NVIDIA Open Agent Safety Platform is designed to put enforceable safeguards around autonomous AI systems, from testing through deployment. It combines runtime software controls with independent hardware monitoring to help keep agents within permitted boundaries.

The NVIDIA Open Agent Safety Platform is designed to put enforceable safeguards around autonomous AI systems, from testing through deployment. It combines runtime software controls with independent hardware monitoring to help keep agents within permitted boundaries.

OpenShell Brings Runtime Security Controls to AI Agents

NVIDIA’s platform brings together OpenShell, an open-source runtime that tracks agent activity and enforces policies, and Sentry, a hardware-based watchdog that can independently monitor and quarantine agents in milliseconds, according to the company. NVIDIA announced the platform on September 28.

NVIDIA says the platform responds to cases where AI agents have bypassed application-layer safeguards while completing assigned tasks. Rather than relying only on the AI model or its agent harness, the approach adds monitoring and controls at the software, hardware and compute layers.

OpenShell creates a runtime security boundary for agents operating on CPUs. It governs access to systems, data and services while keeping a record of agent actions. NVIDIA says the software is now broadly available as open source and can be extended to third-party compute platforms from Arm and Intel.

Its protections include sandboxed execution, controlled access to services, credential management and policy enforcement. A supervisor associated with each OpenShell sandbox checks outbound requests against policy. The sandbox also uses kernel-level controls over filesystem and process activity for the agent workload.

OpenShell can also support human oversight of agent activity. Salesforce, for instance, has integrated it with Slack, allowing users to review activity and audit events and to approve or reject requests for additional permissions.

More than 100 organizations are working with the Open Agent Safety Platform. They include Anthropic, Microsoft, CrowdStrike, Palo Alto Networks, SAP, Salesforce and ServiceNow. NVIDIA also says organizations in financial services, energy and robotics are working with the technology, including on systems that can take action in the physical world.

Sentry Adds Hardware-Based Enforcement for AI Agents

Sentry is an optional hardware layer that uses NVIDIA BlueField-4 data processing units (DPUs) as an out-of-band watchdog. NVIDIA says it continuously monitors agent behavior from an isolated trust domain and can quarantine or stop an agent within milliseconds if it moves beyond its permitted boundaries.

Designed to work independently of the agent, Sentry applies governance controls to requests involving data, tools, application programming interfaces (APIs) and services. This creates a separate enforcement layer outside the agent’s software environment.

NVIDIA says Open Agent Safety Platform software—including OpenShell and related skills—is available through its developer resources and GitHub.

The launch comes as concerns grow the behavior and permissions of autonomous systems. Earlier guidance on agentic AI from the UK’s National Cyber Security Centre (NCSC) warned that agents may have broad access to systems, data and tools. It also noted that rapid autonomous activity can make unexpected behavior more difficult to detect.

Image credit: Sundry Photography / Shutterstock.com

A considerable amount of time and effort goes into maintaining this website, creating backend automation and creating new features and content for you to make actionable intelligence decisions. Everyone that supports the site helps enable new functionality.

If you like the site, please support us on Patreon or Buy Me A Coffee using the buttons below.