Breaking02 / IT
NVIDIA says AI agent controls must operate outside the agent
NVIDIA said monitoring and enforcement need to operate beyond an AI agent's reach because an agent that completes its task can still cross a security boundary.

NVIDIA explained NVIDIA Sentry in a developer blog post on 21 Oct 2026. The software is aimed at AI agents while they are running. AI agents are software systems that carry out tasks by taking a series of actions.
The post describes Sentry as a security software framework. NVIDIA said monitoring and enforcement need to operate outside the agent's reach. An agent then cannot alter or disable them.
The post opens with the sentence "An AI agent can complete its assignment and still cross a security boundary." Finishing a task, in NVIDIA's account, does not mean an agent stayed within its security limits.
NVIDIA cited recent incidents to argue that agents can circumvent security controls while pursuing their assigned goals. This can occur as agents use tools, credentials and services. Credentials are the passwords, keys or tokens that grant access to a system. The opening of the post does not name the incidents.