CrowdStrike, Nvidia launch SafeMind AI security tools

CrowdStrike has announced the launch of SafeMind, a family of artificial intelligence models and agent harnesses designed specifically for security work, in collaboration with Nvidia Corp. SafeMind aims to go beyond simply identifying threats and instead defeat them using AI. The company also unveiled the Cyber Superintelligence Lab, a research organization dedicated to producing these models.

SafeMind includes two models: Red Tempest, which simulates AI-driven adversaries to run advanced attack scenarios, and Blue Solano, which applies containment measures used in live incidents. Both models operate in a closed loop, improving each other's effectiveness. They also drive frontier and open-source models from other providers.

SafeMind will be integrated natively into the CrowdStrike Falcon platform, with standalone access through the Project QuiltWorks program. The models were trained on CrowdStrike's unique cyber dataset, comprising fifteen years of incident response fieldwork, sensor telemetry, and threat intelligence. They were built on Nvidia's Nemotron family and tested against leading frontier models and open-source baselines, achieving a 29% higher detection rate, six times faster remediation, and a 99% reduction in detection and remediation costs.

CrowdStrike's Cyber Superintelligence Lab, led by chief AI and autonomous systems officer Bartley Richardson, brings together AI researchers, offensive operators, and incident responders under one charter. The lab uses telemetry from CrowdStrike's Falcon sensors deployed in customer environments to train the models.