CrowdStrike and NVIDIA launch SafeMind cybersecurity AI platform

CrowdStrike and NVIDIA announced SafeMind, a cybersecurity-focused AI initiative built on NVIDIA’s Nimotron. The system combines specialized models with multiple harnesses designed to reason, call tools, and work through attack-and-defense tasks such as finding vulnerabilities and writing detections.

CrowdStrike said it is deploying SafeMind internally and building a digital twin of its environment for attack-defense simulations. The companies said their goal is to make the system better, faster, and more cost-effective than other open-source and frontier models for cybersecurity use cases; the claimed Arc AGI 3 benchmark improvement from 30% to 100% was presented by the companies and was not independently verified in the discussion.

The SafeMind system includes the Red Tempest offensive model and Blue Solano defensive model. CrowdStrike said both the models and harnesses will be available through its platform, with direct customer access offered through a trusted-access program in Quiltworks. The companies also described a red-blue training loop in which defensive models continually learn from offensive models.