NVIDIA and CrowdStrike have announced a collaboration on the NVIDIA Open Agent Safety Platform, an open-source initiative designed to establish multi-layered security for autonomous AI agents. As AI agents move beyond simple assistants to entities with identities and system access, the security model must shift from model-based safeguards to a full-stack approach. This includes runtime governance and hardware-level monitoring to ensure that agents operate within strictly defined policy boundaries.
The partnership integrates NVIDIA's software and hardware technologies, such as OpenShell and BlueField-4 DPUs, with the CrowdStrike Falcon platform. This combination provides out-of-band monitoring and real-time enforcement, allowing organizations to treat AI agents as privileged identities. By connecting agent activity to broader enterprise telemetry, defenders can gain the necessary context to detect and mitigate machine-speed threats in increasingly autonomous environments.
Top comments (0)