Palo Alto Networks, NVIDIA tighten control over AI agents

Palo Alto Networks already worked with NVIDIA, and this expansion pushes that relationship into a specific problem: AI agents that can write code, retrieve company data and use software tools with limited human input. Because those agents act rather than just answer, the security question shifts from what a model says to what an agent is allowed to reach.

The body lays out several distinct pieces rather than one product. Prisma AIRS AI Runtime Security on NVIDIA BlueField data processing units runs security controls on a processor separate from the host, so network policy can be enforced independently of the application running the agent. NVIDIA DOCA Argus can monitor agent execution paths and tool calls, enforce access rules for networks and external APIs, and send telemetry to Cortex XSIAM for analysis and response. A planned OpenShell and Idira integration would add identity and secrets management for agent sandboxes.

The stakes appear to hinge on timing and maturity rather than ambition. OpenShell and the BlueField deployment are generally available, while the Vera deployment is described as forward-looking, so how quickly companies can standardize on this design may depend on when that future architecture arrives and how the planned Idira integration lands.