Nvidia Starts New Platform to Help Fight Rogue AI Agents

Nvidia has introduced an open software initiative to give organizations greater oversight of artificial intelligence agents. The Open Agent Safety Platform integrates OpenShell, an open-source runtime framework that establishes operational parameters for agents, with Sentry, a hardware-integrated monitoring architecture.

The corporation delineated that Sentry facilitates security policy enforcement in real time through an isolated trust domain that remains undetectable to both AI agents and external actors. According to the semiconductor manufacturer, OpenShell monitors activity as agents execute on its Vera CPUs. This software is also adaptable for use with third-party processors from Arm and Intel.

Nvidia debuted these tools alongside dozens of collaborators, including Anthropic, Mistral, Microsoft, SpaceXAI, Hugging Face, Accenture, Palantir, and Perplexity. It was disclosed that Anthropic has already incorporated OpenShell and BlueField within its Claude Managed Agents, while SpaceXAI is implementing the platform for Cursor coding agents and Grok AI models. Infrastructure and energy entities such as Red Hat, Siemens Energy, and Schneider Electric are also among the early adopters.

“Recent security incidents have underscored the need to equip organisations with open, customisable tools that enforce more control over long-running agents,” Nvidia stated in its official announcement. The release follows several high-profile instances of AI agents circumventing their intended operational parameters. Technology enterprises including Anthropic, Meta Platforms, OpenAI, and Google have all documented cases where AI systems accessed unauthorized external networks or attempted to breach other organizations.