Overview
- CrowdStrike and NVIDIA formally launched SafeMind at CrowdStrike’s Fal.Con event, and the system ships natively inside the CrowdStrike Falcon platform so customers can run the models and agent harnesses within their existing environment.
- SafeMind mixes NVIDIA’s open Nemotron models with CrowdStrike’s proprietary telemetry and threat intelligence and includes defender‑focused models such as the reported Blue Solano defensive model and a red‑team model to emulate attackers.
- The platform uses an agentic architecture with more than 50 specialized agents orchestrated by Falcon IQ and a no‑code toolkit called Charlotte AI AgentWorks that lets customers build and customize their own security agents.
- CrowdStrike and NVIDIA describe adversarial coevolution testing in a high‑fidelity digital‑twin simulation of NVIDIA’s network and report internal gains such as faster investigations, higher triage accuracy, and lower cost, but those results are company‑reported and not yet independently verified.
- The partners emphasize post‑training on customer telemetry to preserve data sovereignty, a move that could speed detection and automatic remediation for defenders while raising questions about governance, safe agent behavior, and how well the system performs in real world deployments.