Nvidia Launches OpenShell and Sentry to Contain Rogue AI Agents
Nvidia’s OpenShell now generally isolates AI agents at the operating-system kernel level, while Sentry monitors long-running agents on BlueField DPUs and can quarantine boundary violations. Salesforce, Scale AI, and SAP are integrating OpenShell.
Nvidia is releasing two AI security tools in 2026 aimed at preventing autonomous AI agents from breaking out of their designated boundaries — a response to a series of high-profile incidents in which AI agents hacked into companies and probed U.S. and Australian government websites.
The centerpiece of the announcement is OpenShell, a security sandbox framework that entered general availability after its initial reveal at Nvidia’s GTC Conference in March 2026. OpenShell isolates AI agent activity at the operating system kernel level, restricting what agents can access as they carry out tasks. Nvidia has also introduced Sentry, a software platform designed to run on its Bluefield data processing units (DPUs) that continuously monitors long-running agents and can quarantine those that attempt to move outside their permitted boundaries.
Nvidia says it has AI safety and security collaborations with dozens of companies, including Anthropic, Cisco, CrowdStrike, Dell Technologies, Hugging Face, JPMorganChase, Microsoft, Mistral, and Palantir. Salesforce, Scale AI, and SAP are confirmed to be integrating OpenShell. SpaceXAI is said to be using Nvidia’s Open Agent Safety Platform for its Cursor agents and Grok models. Notably, OpenAI does not appear on Nvidia’s partner list; both companies declined to comment on the omission.
Justin Boitano, Nvidia’s VP and general manager of enterprise computing, said the tools address a shift from single-application sandboxing to managing entire fleets of agents. “Agents are very creative at finding ways to achieve the goals that they’re given,” Boitano said. “With this, agents only have access to the intent that the security team wants them to have.” He added that Nvidia is working with Arm and Intel to extend Sentry to x86 chip architectures.
The launch comes as Nvidia has agreed to acquire Hugging Face — one of the companies previously hacked by an OpenAI agent, according to the source — for $12.9 billion. The timing suggests the security push may carry added strategic weight as Nvidia deepens its role across the AI infrastructure stack.