CrowdStrike, Nvidia launch SafeMind security models

CrowdStrike's Fal.Con event arrived with a stark backdrop: AI has compressed the time an attacker needs to move from an initial foothold to another target, a measure called "breakout time." CrowdStrike's 2026 Global Threat Report put the average eCrime breakout time at 29 minutes, with the fastest attack at 27 seconds. CrowdStrike CEO George Kurtz has presented this as an escalating annual trend at past Fal.Con keynotes, and Adam Meyers, senior vice president of intelligence, said the company tracked 26 agentic adversaries in the last 30 days — more than it tracked in the year before.

SafeMind is CrowdStrike's answer. Built with Nvidia on its Nemotron family and trained on CrowdStrike incident data, it splits into two models: Red Tempest, which scans a digital twin of a customer's environment for attack paths, and Blue Solano, which writes rules to block them. Nvidia's Justin Boitano described this as an iterative loop that hardens the environment.

The pitch extends beyond the models themselves. CrowdStrike is pairing SafeMind with partnerships — Project QuiltWorks, a cross-industry alliance of technology companies, law firms and consultancies, and a long-running tie-up with Amazon Web Services. Whether that proves enough hinges on how quickly AI-based threats keep outpacing human security teams, and on whether CrowdStrike's defender-focused models can hold an advantage once attackers gain access to the same capabilities.