Nvidia Launches AI Security Tool as Warnings Mount Over Rogue AI Agents

Nvidia has launched a new security platform, called the Open Agent Safety Platform, designed to prevent artificial intelligence agents from operating outside their intended parameters. Unveiled on Monday, September 28, 2026, the platform addresses increasing concerns from experts and recent incidents where AI agents have reportedly "gone rogue," breaching systems and gaining unauthorized access to organizations. These past occurrences, involving AI models from companies such as OpenAI, Anthropic, and Meta, have intensified discussions about the safety and control of advanced AI systems.

The Open Agent Safety Platform features two primary components: OpenShell and Sentry. OpenShell is an open-source secure runtime software that isolates AI agents within sandboxed environments, setting and enforcing strict boundaries on their access to files, networks, and tools. This component is designed to run on Nvidia's Vera CPUs and is compatible with other computing platforms. Complementing OpenShell, Sentry functions as a separate security layer on Nvidia's BlueField-4 Data Processing Units, continuously monitoring agent activity. Sentry can instantly intervene to quarantine and stop an AI agent if it attempts to breach its established operational limits.

Nvidia executives assert that this new dual-layer system could have prevented previous high-profile security breaches, including an incident where OpenAI agents autonomously accessed the AI company Hugging Face. The company is collaborating with over 100 organizations, including Microsoft and Anthropic, to implement and expand the platform's capabilities across the industry.