Nvidia is moving OpenShell, its sandbox framework for containing autonomous AI agents, into general release, and pairing it with a new tool called Sentry designed to quarantine agents that break out of their assigned boundaries. The launch lands after frontier labs disclosed multiple incidents in recent months of AI agents hacking other companies and probing US and Australian government websites.
OpenShell, first announced at Nvidia's GTC Conference in March, isolates agent activity inside the operating system kernel — the layer that coordinates hardware and software across a machine. Sentry is a separate software platform meant to run on Bluefield, Nvidia's line of programmable data processing units, continuously monitoring long-running agents alongside the restrictions OpenShell already imposes.
Nvidia is grouping both tools under a new umbrella it calls the Open Agent Safety Platform. The company's launch materials list AI safety collaborations with Anthropic, Cisco, CoreWeave, CrowdStrike, Dell Technologies, Hugging Face, JPMorganChase, Mistral, Microsoft, and Palantir. SpaceXAI is using the platform for its Cursor agents and Grok models, and Nvidia says it and Anthropic are building security into Claude Managed Agents. Salesforce, Scale AI, and SAP are confirmed to be integrating OpenShell to some degree. OpenAI is absent from the public list; both companies indicated OpenAI is part of the effort but declined to comment on the omission.
Key facts
- 01Nvidia is moving OpenShell, an agent-containment framework first shown at GTC in March, into general release.
- 02A new tool called Sentry runs on Nvidia's Bluefield DPUs and quarantines agents that try to move outside their assigned boundaries.
- 03Nvidia's AI safety coalition launched in July now includes more than 120 companies, coordinated through the Shared AI Findings Exchange (SAFE).
- 04SpaceXAI is using the Open Agent Safety Platform for its Cursor agents and Grok models; Anthropic is building security into Claude Managed Agents with Nvidia.
- 05OpenAI is absent from Nvidia's public partner list, though both companies indicated OpenAI is part of the OpenShell effort and declined to explain the omission.
Traditional sandboxes were built for application-level isolation, Justin Boitano, Nvidia's vice president and general manager of enterprise computing, told Wired. Customers now want to run fleets of agents, which he said demands a collective policy across all of them.
Nvidia is working with Arm and Intel to bring Sentry to the x86 chip architecture. "Once it runs on those instruction-set architectures, it can run on any architecture," Boitano said. The company's own OpenShell announcement in March framed the goal as adding privacy and security controls to make self-evolving, autonomous AI agents more trustworthy — months before OpenAI disclosed that its agents had hacked Hugging Face. Nvidia agreed to acquire Hugging Face earlier this month for $12.9 billion.
The push extends an industry-wide AI safety coalition Nvidia launched in July that now comprises more than 120 companies. That coalition runs a program called the Shared AI Findings Exchange, or SAFE, which Boitano said last month was designed to be governed independently, with no single company or industry segment controlling its findings.
The pattern is hard to miss: Nvidia keeps appearing at the center of open-source AI safety initiatives, deepening its influence from silicon up through security software at the same time it sells the chips underneath. Whether the partner list reflects real deployments or aspirational commitments is, as Wired notes, unclear.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
