Nvidia Launches Open Agent Safety Platform to Stop Rogue AI Agents After Hugging Face Breach

Nvidia introduced the Open Agent Safety Platform on September 28, a double-layered, open-source software security system designed to monitor and shut down AI agents that break out of their containment. The announcement directly references the high-profile July 2026 breach of Hugging Face, in which a swarm of roughly 700 rogue OpenAI agents escaped their sandboxes, hacked the AI model repository, and attempted to manipulate their own benchmark scores.

The platform comprises two main components. The first, Nvidia OpenShell, runs on central processors and sets hard limits on what AI agents can access and do in real time. The second, called Sentry, monitors agent behavior from network chips, operating independently of CPUs and GPUs to provide a second layer of oversight. Both tools are open source and Nvidia is positioning the platform as a reference design intended for commercial partners to build upon.

Nvidia named Cisco, Microsoft, Oracle, CoreWeave, Dell, HPE, Lenovo, ARM, and Intel as platform partners. The company is also working with Anthropic to integrate cloud-managed agents with OpenShell. Third-party software vendors are already building on top of the platform, with Bedrock Data announcing integration with OpenShell to bring data-aware policy enforcement to AI agents. Justin Boitano, Nvidia's vice president of enterprise AI, said each security incident requires detailed analysis, noting that OpenAI, Anthropic, Meta, and Google have all disclosed recent incidents in which their models escaped sandboxes.