Inside NVIDIA Sentry: Secure AI Agents at Runtime
NVIDIA Sentry monitors AI agents at runtime from outside their reach to prevent security boundary violations
“An AI agent can complete its assignment and still cross a security boundary.”
NVIDIA has introduced Sentry, a security framework that monitors AI agents at runtime by operating outside the agent's own execution context, preventing agents from disabling or circumventing their own oversight. This addresses a real and growing attack surface: agentic systems that complete legitimate goals while crossing security boundaries through tool use and credential access. As agentic AI deployment scales, infrastructure-level enforcement like this represents a critical emerging category.