CrowdStrike and Nvidia Corp. used the Fal.Con 2026 keynote in Las Vegas on Tuesday to launch CrowdStrike SafeMind, a family of security-specific AI models and agentic harnesses built on Nvidia Corp.'s Nemotron architecture1,2,5. The system pairs two purpose-built models — Red Tempest, which probes for attack paths, and Blue Solano, which works to close them — inside CrowdStrike's Falcon platform10.
"We're at an inflection point in cybersecurity," Jensen Huang told the audience. "Attacks are now automated. Defense has to be, too"9.
How SafeMind works
SafeMind is trained on Falcon telemetry, CrowdStrike threat intelligence, and 15 years of incident-response data, according to CrowdStrike. CSO reports that the system creates a digital twin of an enterprise environment, allowing the red-team model to test attack paths while the blue-team model responds. CrowdStrike describes SafeMind as built for security work rather than general use.
CrowdStrike's vendor-reported benchmarks claim a 29% higher detection rate, 6x faster end-to-end remediation, and 99% cost savings on detection and remediation compared with leading frontier models and open-source baselines. Those figures are self-reported.
A new research unit, the Cyber Superintelligence Lab, produced the SafeMind models. CrowdStrike also extended its Falcon platform across Google Cloud's enterprise AI ecosystem as a separate integration.
The Nvidia Corp. angle
Jensen Huang framed cybersecurity as AI's next major market3,4. George Kurtz, CrowdStrike's CEO and founder, shared the stage for the announcement. The partnership centers on Nvidia Corp.'s Nemotron serving as the base architecture for SafeMind's models.
ANALYSIS The joint launch marks a concrete instance of Nvidia Corp. lending its model stack to a vertical software vendor for domain-specific fine-tuning, rather than selling only compute. By splitting SafeMind into adversarial (Red Tempest) and defensive (Blue Solano) components operating inside a digital twin, CrowdStrike is positioning the product as a continuous red-team/blue-team loop that runs without human-in-the-loop latency. The vendor-reported performance claims — particularly the 99% cost-savings figure — will face scrutiny until third-party evaluations confirm them in production environments with realistic false-positive rates.