Nvidia is launching a security platform to stop rogue AI agents from hacking systems

Over the past few months, frontier AI labs have revealed numerous instances of AI agents infiltrating other companies or probing official government websites in the US and Australia. Nvidia has been a leading force in promoting open source security for AI, and now it is introducing a new software security platform called OpenShell, along with making an agentic AI sandbox more widely available.

OpenShell, first announced at Nvidia's GTC Conference in March, is a framework designed to contain AI agents as they perform tasks and isolate their activity within the operating system kernel. Nvidia has partnerships with a wide range of tech companies, including Anthropic, Cisco, CoreWeave, CrowdStrike, Dell Technologies, Hugging Face, JPMorgan Chase, Mistral, Microsoft, and Palantir, all of which are integrating OpenShell to some degree.

However, OpenAI is notably absent from Nvidia's list of partners. Despite this, both companies have indicated that OpenAI is part of the OpenShell effort, but declined to comment on the reason for its exclusion. Nvidia has also developed Sentry, an isolated security domain for chips that continuously monitors long-running AI agents. This tool, meant to run on Nvidia's Bluefield programmable data processing units (DPUs), can act as an independent mechanism to quarantine agents attempting to move outside their boundaries.

Justin Boitano, Nvidia's vice president and general manager of enterprise computing, explains that while traditional sandboxes are built for application-level isolation, the need to run fleets of agents calls for a collective policy across all agents. Boitano also mentions that Nvidia is working with Arm and Intel to create a version of Sentry that works on the x86 chip architecture, allowing it to run on any architecture.

Nvidia has positioned these efforts as part of a new open-source framework called the Open Agent Safety Platform, which now includes both OpenShell and Sentry. The company is leading an industry-wide AI safety coalition with over 120 companies, aiming to reduce AI risks and create a Shared AI Findings Exchange (SAFE) governed independently.