CrowdStrike and Nvidia push AI agent security down to the silicon

CrowdStrike and Nvidia are expanding protections for autonomous AI agents from the software layer into the hardware itself. The bet is that agents moving at machine speed need guardrails that move just as fast. What the companies are actually building The core of the expansion pairs Nvidia’s DOCA Argus telemetry with its BlueField-4 Data Processing Units, or DPUs. DPUs are specialized processors that sit in the infrastructure layer. Running security monitoring there means watching activity from beneath the software, rather than relying only on tools that run alongside it. CrowdStrike plugs that infrastructure-level data into its Falcon platform. The goal is to track what AI agents do and how data moves inside what Nvidia calls “AI factories,” the large-scale environments where enterprises build and run AI workloads. The most recent milestone landed on October 20, 2026. That’s when the companies unveiled advanced integrations built around infrastructure-level security signals, which are designed to follow agent activity across networks. Nvidia’s Open Agent Safety Platform, as of October 2026, also includes additional telemetry and quarantine capabilities. A partnership that has been...