NVIDIA and CrowdStrike announce SafeMind for agentic cyberdefense

NVIDIA CEO Jensen Huang and CrowdStrike CEO George Kurtz introduced CrowdStrike SafeMind at Fal.Con 2026 in Las Vegas, positioning the system as a response to automated cyberattacks. SafeMind was developed by the CrowdStrike Cyber Superintelligence Lab and ships natively in the CrowdStrike Falcon platform.

SafeMind combines CrowdStrike’s beyond frontier-capable models and proprietary agentic harnesses with defensive models built on NVIDIA Nemotron. CrowdStrike post-trained Nemotron open models on its cyber experience and threat data, pairing offensive and defensive AI in a continuous coevolution loop designed to harden customer environments. NVIDIA Nemotron 3 Ultra orchestrates the defensive agent harness, while a fine-tuned Nemotron 3 Super powers SafeMind’s rule-generation sub-agent.

CrowdStrike said internal evaluations showed the Blue Solano model, based on Nemotron 3 Super, delivered higher accuracy rates than leading frontier models at 99% lower cost. The company also cited AI-enabled attacks rising 89% in the past year and the fastest eCrime breakout time reaching 27 seconds, underscoring its case for machine-speed defense.

NVIDIA tested SafeMind models and harnesses in a simulated digital twin of its network, using a red-team agent to find exploits and a blue-team agent to close them. CrowdStrike also announced Falcon IQ, powered in part by NVIDIA Nemotron models, using more than 50 agents to automate workflows across assessment, prioritization and remediation.