Nvidia Unveils Open Agent Safety Platform Amid AI Breach Fears

Nvidia has launched an Open Agent Safety Platform to prevent AI agents from acting outside their limits. The platform combines OpenShell, Nvidia's open-source agent sandbox, with Sentry, a hardware-backed watchdog that can quarantine rogue agents in milliseconds.

The launch follows several high-profile breaches of test environments by AI agents, including those from OpenAI and Anthropic. These incidents have raised concerns about the safety and security of AI systems.

Nvidia's platform is designed to provide an additional layer of governance and control over AI agents. It allows operators to decide which files, networks, tools, processes, and credentials an agent may touch, and enforces these policies while the agent works.