CrowdStrike & NVIDIA Launch Open Agent Safety Platform to Secure AI Stack

What Happened

CrowdStrike and NVIDIA announced the NVIDIA Open Agent Safety Platform, an open reference design that embeds security controls directly into the agentic stack of autonomous AI systems. The design builds on Falcon’s agentic SOC framework, adding policy enforcement, runtime monitoring, and boundary checks for AI agents.

Impact

Organizations running AI workloads on NVIDIA hardware—such as Jetson, A100, or DGX systems—will now have a standardized security baseline. The platform enables tighter isolation between AI models and host resources, reducing the attack surface for adversaries that target model integrity or data leakage.

What Defenders Should Do

• Review the reference design and map it to your existing AI deployment architecture.
• Update Falcon to the latest version that supports the Open Agent Safety Platform hooks.
• Implement policy rules that enforce model isolation, input validation, and output monitoring.
• Configure runtime telemetry to surface agentic activity in Falcon’s XDR console.
• Validate that all AI agents run with the minimum required privileges and that firmware updates are signed.

Next Steps

Defenders should schedule a security review of their AI pipelines, test the reference design in a staging environment, and coordinate with NVIDIA’s support for any hardware‑specific integration questions. Regularly monitor Falcon’s threat intelligence feeds for emerging AI‑related threats and adjust policies accordingly.