CrowdStrike and Nvidia are extending protection for autonomous AI agents to the hardware layer by combining Nvidia’s DOCA Argus telemetry and BlueField-4 processors with CrowdStrike’s Falcon platform. These components monitor activity beneath the software layer to track agent actions and data movements across large-scale AI factories. On October 20, 2026, the companies announced infrastructure-level security integrations, alongside additional telemetry and quarantine capabilities in Nvidia’s Open Agent Safety Platform. The SafeMind models, introduced in September 2026, reportedly deliver a 29% higher detection rate and six times faster remediation than previous models, although these gains still require validation in real-world deployments.
Source: Read the original article
