Miles Jamison 
- OpenShell creates a runtime boundary for AI agents and tracks their actions while enforcing policies
- Sentry runs on NVIDIA BlueField-4 DPUs to quarantine misbehaving agents in milliseconds
- More than 100 organizations are working with NVIDIA’s agent safety technologies
NVIDIA has introduced an open software offering and reference system design intended to govern and secure autonomous artificial intelligence agents from testing through deployment.
What Does NVIDIA’s Agent Safety System Include?
The company said Monday the NVIDIA Open Agent Safety Platform combines OpenShell secure runtime software with the Sentry reference system design. The technologies are designed to establish control over the software, hardware, compute and robotics systems used by AI agents.
Now available through NVIDIA’s developer resources page and GitHub, OpenShell establishes a runtime boundary for agents operating on CPUs and traces their actions while enforcing policies. The open-source software can also be extended to work with computing hardware from Arm and Intel.
Sentry operates on NVIDIA BlueField-4 data processing units and independently monitors agent activity. NVIDIA said the system can quarantine an agent that attempts to move outside its software boundary within milliseconds.
Founder and CEO Jensen Huang said AI safety must be solved for the technology to deliver its potential to society.
“Safety and security require full-stack engineering,” Huang added.
How Does NVIDIA Work With Other Organizations?
NVIDIA said more than 100 organizations are working with its agent safety technologies.
Anthropic is working with NVIDIA on additional security controls for its Claude Managed Agents. Scale AI is incorporating the platform into agentic infrastructure for enterprise and government customers, while SpaceXAI uses it for Cursor coding agents and Grok models.
SAP is building OpenShell into Joule Studio and contributing engineering work to the open-source software. Salesforce has integrated OpenShell with Slack to let teams monitor agent activity and audit events and approve or reject requests for additional permissions.
Figure, Gecko Robotics, Skild AI and other robotics companies are also using OpenShell for autonomous systems operating in the physical world.
Other organizations working with NVIDIA’s agent safety technologies include Accenture, Cisco, CrowdStrike, Dell Technologies, Deloitte, HPE, Hugging Face, JPMorganChase, Microsoft, Palantir, Palo Alto Networks, Perplexity, Red Hat, Siemens and ServiceNow.
What Other AI Efforts Has NVIDIA Pursued?
NVIDIA has used a similar partner-based approach in other AI efforts. In 2025, the company worked with Booz Allen Hamilton, Cisco, MITRE, ODC and T-Mobile to develop an AI-native wireless stack for 6G that combined NVIDIA software with partner technologies.
The company developed computing technologies for autonomous spacecraft and other edge environments, including systems designed to process sensor, navigation and vision data onboard spacecraft.
