Nvidia is making OpenShell, its sandbox for containing AI agents, generally available while introducing Sentry, a separate software layer intended to monitor and quarantine agents running on Nvidia hardware. The tools form part of the company’s broader Open Agent Safety Platform, according to Nvidia and reporting by WIRED.
A sandbox for agents
OpenShell was first announced at Nvidia’s annual GTC conference in March. The framework is designed to isolate an agent’s activity in the operating system kernel, limiting the parts of a computer it can access while it performs assigned tasks. Its general release gives users access to the containment framework, although the source does not specify a licensing scope beyond describing Nvidia’s effort as open source.
Monitoring beyond the sandbox
Sentry is intended to provide an additional security boundary for long-running agents. Nvidia describes it as an isolated security domain that continuously monitors agent activity and can quarantine an agent that moves beyond its permitted boundaries. The software is designed for deployment on Nvidia’s BlueField programmable data processing units, so its stated role depends on specific infrastructure rather than OpenShell alone.
Nvidia says it is working with Arm and Intel on a version of Sentry for the x86 architecture. That work is described as ongoing, and the source does not report an availability date for the broader architecture support.
Industry backing remains uneven
Nvidia says companies including Anthropic, Cisco, CoreWeave, CrowdStrike, Dell Technologies, Hugging Face, JPMorganChase, Mistral, Microsoft and Palantir are involved in related safety collaborations. It also says Salesforce, Scale AI and SAP are integrating OpenShell to some degree. The reported status of Nvidia’s wider partner list is less clear, and the source does not establish that every named company has adopted the tools.
OpenAI is notably absent from Nvidia’s announcement even though both companies have indicated that the AI lab is part of the OpenShell effort. Neither company explained the omission, according to WIRED.
Nvidia’s larger role
The release extends Nvidia’s involvement in industry-wide AI safety initiatives. The company launched a coalition in July that it says now includes more than 120 companies and is working on the Shared AI Findings Exchange. By combining agent controls with its hardware and infrastructure position, Nvidia is also positioning itself to influence how security policies for autonomous software are implemented across the industry.
AI-generated editorial illustration; not a documentary photograph.
Source: Nvidia’s Answer to Rogue Agents Is an Open-Source AI Security System
SIGNAL TITANS
Leave a Reply