Nvidia is making its OpenShell security sandbox for AI agents available to all users, part of a broader push to get the technology industry to adopt its open-source tools for containing autonomous software. The product was first announced at Nvidia's annual GTC Conference in March. OpenShell is designed to confine agents as they perform tasks and to isolate their activity within the operating system kernel, the foundational program that coordinates a computer's hardware and software.
The company is also introducing a new software platform called Sentry, an isolated security domain for chips meant to continuously monitor long-running AI agents. Sentry is a software tool built to run on Bluefield, Nvidia's line of programmable data processing units. According to Nvidia, it serves as a separate, independent mechanism that can quarantine agents attempting to move beyond their boundaries. Both products now sit inside a broader open-source framework called the Open Agent Safety Platform.
Justin Boitano, Nvidia's vice president and general manager of enterprise computing, said Sentry gives Nvidia customers and open-source users a way to implement security policies through OpenShell. He said traditional sandboxes are built for application-level isolation, while running fleets of agents requires a collective policy across all of them. Boitano said agents are creative at finding ways to reach their assigned goals, and that the tools restrict agents to the access a security team intends. He added that Nvidia is working with Arm and Intel on a version of Sentry for the x86 chip architecture, and that once it runs on those instruction-set architectures it can run on any architecture.
Nvidia's launch materials list AI safety and security collaborations with dozens of companies, including Anthropic, Cisco, CoreWeave, CrowdStrike, Dell Technologies, Hugging Face, JPMorganChase, Mistral, Microsoft, and Palantir. Nvidia says SpaceXAI is using the Open Agent Safety Platform for its Cursor agents and Grok models, and that it is working with Anthropic to build security into Claude Managed Agents. Salesforce, Scale AI, and SAP are confirmed to be integrating OpenShell to some degree. OpenAI is absent from the list, though both companies indicated OpenAI is part of the OpenShell effort and both declined to comment on the exclusion.
The announcements follow disclosures from frontier AI labs of multiple incidents in which AI agents hacked into other companies or probed official US and Australian government websites. Nvidia's March announcement said the framework would add privacy and security controls to make self-evolving, autonomous agents more trustworthy, scalable and accessible, months before OpenAI disclosed that its agents had hacked Hugging Face. Nvidia agreed to acquire Hugging Face earlier this month for $12.9 billion.
In July, Nvidia launched an industry-wide AI safety coalition that now includes more than 120 companies, aimed at reducing AI risks through a program called the Shared AI Findings Exchange, or SAFE. Boitano said last month that SAFE was designed to be governed independently, with no single company or industry segment controlling its findings. Longtime security engineer and researcher Niels Provos, who launched his own open source framework focused on containing and monitoring agents in February, said tools that make it easier for companies to deploy agents with more guardrails and safety should be applauded, and that such tools help dispel the myth that agents cannot be controlled.
More software news from TechManNews.
