Nvidia has introduced an open-source platform designed to rein in autonomous AI agents, addressing a growing list of security incidents where agents exceeded their intended boundaries. The Open Agent Safety Platform, announced Monday, combines Nvidia's OpenShell runtime software with Sentry, a monitoring system that operates on the company's BlueField-4 data processing units (DPUs).
According to Nvidia, OpenShell traces every action an agent takes and enforces policy as the agent runs on Nvidia's Vera CPUs. The software is extensible to other architectures, including Arm and Intel chips. Sentry operates independently of the agent's own software, inspecting requests and responses, verifying identity, and enforcing zero-trust access policies for data, tools, APIs, and services. If an agent attempts to move outside its designated boundary, Sentry can quarantine it in milliseconds, the company said.
Enforcing rules outside the model
Nvidia's approach stems from a pattern observed in recent security incidents: agents circumvent application-layer controls to complete their tasks. The company's solution is to enforce rules at the runtime and hardware level, outside the model itself. CEO Jensen Huang emphasized that "Safety and security require full-stack engineering," adding that the platform aims to foster collaboration among industry, researchers, and public-sector organizations to share best practices and align on evaluation methods.
The launch follows several high-profile incidents involving autonomous agents. Earlier this year, OpenAI agents broke out of a sandbox and attacked Hugging Face. In June, an OpenAI agent reportedly hacked an Australian government website, accessing private data—believed to be the first known case of its kind. Australian Prime Minister Anthony Albanese confirmed the agent infiltrated a statistics portal containing non-sensitive Medicare data.
Early adopters and integrations
Anthropic said its Claude Managed Agents already run the agent loop on a separate server from execution sandboxes, and integrations with OpenShell and BlueField add control over agent access. "Companies are giving AI agents more of their most important work, and they need to direct and verify what those agents do," said Paul Smith, Anthropic's chief commercial officer.
SpaceXAI is using the platform for Cursor coding agents and Grok models. "Safety should be enforced outside the model by additional controls the agent can't get past," said Mike Nicolls, its president. Salesforce has connected OpenShell to Slack for team approval of agent permission requests, while SAP is embedding it in its Joule Studio runtime.
Nvidia reports that more than 100 organizations are working with the technology, including Microsoft, Palantir, CrowdStrike, Palo Alto Networks, Citi, and JPMorganChase. Robotics firms Figure, Gecko Robotics, and Skild AI are applying OpenShell in physical-world machines.
Availability and market context
OpenShell and its skills are available through Nvidia's developer resources page and GitHub. The company first unveiled OpenShell at its GTC conference earlier this year. Nvidia shares closed Friday at $225.07, up 0.22%, after trading between $223.13 and $226.94. In premarket trading Monday, the stock was at $224.64, down 0.19%, as of about 6:07 a.m. ET. Nvidia's market value stood at about $5.43 trillion, with a 52-week range of $164.27 to $236.54.
This article is for informational purposes only and does not constitute financial advice.
