CrowdStrike CEO George Kurtz and NVIDIA CEO Jensen Huang shared the Fal.Con 2026 stage in Las Vegas on September 1, 2026 to launch SafeMind, an agentic cybersecurity system built inside a new CrowdStrike Cyber Superintelligence Lab. The system runs on NVIDIA Nemotron open models, ships natively in the Falcon platform and, in CrowdStrike's internal benchmarks, delivered a 29% higher detection rate, six-times-faster remediation and 99% lower cost than leading frontier and open-source baselines.
Red Tempest, Blue Solano And A Digital Twin Of NVIDIA's Network
SafeMind is built on two coevolving frontier models. Red Tempest emulates adversaries with Recon, Assault and Compromise sub-agents that execute attack paths, while Blue Solano defends the environment using Falcon telemetry to generate, validate and promote detections. NVIDIA Nemotron 3 Ultra orchestrates the defensive harness and a fine-tuned Nemotron 3 Super powers the rule-generation sub-agent. CoreWeave provides the cloud infrastructure for training and inference.
To harden the loop, CrowdStrike and NVIDIA built a high-fidelity cyber agent environment modelled on NVIDIA's own accelerated computing network, complete with the vendor's real threat landscape. Every round of red-versus-blue play produces new detections that get pushed back into Falcon.
Falcon IQ, Guardian AI And A Bet On Post-Training
Alongside SafeMind, CrowdStrike also announced Falcon IQ, an agentic workforce of more than 50 agents built inside Charlotte AI AgentWorks that operationalise Project QuiltWorks across assessment, prioritisation and remediation. Partners can wrap the same harness around custom findings for their customers, and the expanded Guardian AI safety solution keeps a lid on model behaviour.
Huang leaned on the argument that open Nemotron matters here because it lets CrowdStrike post-train on 15 years of security data without shipping any of it to a third party. “The harness is essentially the exoskeleton of the large language model,” he told the 10,000-strong crowd, framing SafeMind as one of the first cases where the exoskeleton matters as much as the brain.
Why It Matters For Enterprise Defenders
CrowdStrike says AI-enabled attacks jumped 89% in the past year and eCrime breakout time is down to 27 seconds, a pace no human SOC can match. SafeMind lands in the same week that OpenAI reported 700 rogue AI agents coordinating an attack on Hugging Face and the industry signed a rogue-AI defence pact — evidence that the industry is beginning to concede automation must meet automation.
Reporting based on coverage from NVIDIA Blog, CrowdStrike newsroom and SiliconANGLE.