The tech giant, Nvidia has launched an open-source safety platform to help businesses control, monitor and secure autonomous AI agents.
Nvidia Takes Aim at Rogue AI Agents
Nvidia has launched a new platform to make AI agents safer and easier to control.
The Nvidia Open Agent Safety Platform is free and open source. It gives organisations tools to monitor and secure autonomous AI agents.
The platform also helps companies control the computing resources behind these agents. Nvidia built it on its open-source OpenShell software.
The move comes as concerns grow over AI agents acting beyond their assigned limits. Recent security incidents have shown how quickly autonomous systems can become difficult to control.
OpenShell Adds Another Layer of Protection
Nvidia says traditional AI guardrails may not be enough for autonomous agents.
Its OpenShell software creates isolated environments for running groups of AI agents. It also lets organisations track their actions and enforce security rules.
The system works outside the AI model itself. This means an agent cannot easily bypass controls through its own software instructions.
OpenShell supports Nvidia Vera CPUs. It also works with CPUs from Intel and Arm. This approach gives companies more control over what their AI agents can access and do.
Nvidia Sentry Can Stop Agents
Another key part of the platform is Nvidia Sentry. The system uses Nvidia’s DOCA software and runs on BlueField-4 data processing units.
Sentry monitors AI agents in the background. It can inspect requests, check identities and verify actions.
More importantly, it can stop an agent if it crosses a set boundary.
Nvidia CEO Jensen Huang said safety must become a major part of AI development.
“Safety and security require full-stack engineering,” Huang said.
He added that the platform aims to bring companies, researchers and governments together around AI safety.
Big Tech and Robotics Firms Join In
Several companies are already using Nvidia’s safety tools. SpaceXAI says it uses the platform to secure Cursor’s coding agents and the Grok AI model.
“As customers rely more on agents to get real work done, safety should be enforced outside the model,” said SpaceXAI President Mike Nicolls.
Salesforce has also integrated OpenShell with Slack. The integration gives users approval controls, visibility and audit tracking.
Meanwhile, SAP uses the software to improve runtime security for its autonomous AI systems.
Robotics firms, including Figure AI, Gecko Robotics and Skild AI, also use OpenShell. Nvidia’s platform is now available through its developer resources and GitHub. Its real-world reliability at massive scale remains to be tested.
However, the launch shows how seriously the industry now treats AI agent.
