CrowdStrike has established a Cyber Superintelligence Lab dedicated to frontier artificial-intelligence research for cyber defence and AI safety, alongside SafeMind, an agentic security system developed with NVIDIA. NVIDIA will serve as an AI compute design partner and invest $100 million in the lab over five years.
The research organisation brings together CrowdStrike’s AI researchers, offensive-security operators, threat-intelligence specialists and incident responders. It is led by Bartley Richardson, CrowdStrike’s chief AI and autonomous systems officer, and will use security telemetry, analyst-labelled threat data and high-fidelity cyber ranges derived from the Falcon platform.
SafeMind is the lab’s first announced system. It combines two families of specialised models: Red Tempest for offensive security work and Blue Solano for defensive operations. Runtime harnesses provide the models with context, memory, tools, permissions, model routing, orchestration and safety controls, enabling them to carry out extended security workflows instead of producing isolated recommendations.
In SafeMind’s operating loop, red-team agents emulate an attacker and search for exploitable paths. Blue-team agents monitor telemetry, identify weaknesses, generate candidate detections, test them and promote validated protections. The offensive agents then attack the updated environment again. Training and testing take place in cyber ranges that reproduce enterprise conditions without exposing production systems.
The defensive stack uses NVIDIA Nemotron open models. Nemotron 3 Ultra orchestrates the defensive-agent harness, while a fine-tuned Nemotron 3 Super supports the rule-generation sub-agent. CrowdStrike said an internally evaluated Blue Solano model produced higher accuracy than the general-purpose frontier models it tested at 99% lower cost. That figure is a company benchmark rather than an independent comparison. Technical details of the Nemotron implementation and the red-versus-blue testing environment are set out in NVIDIA’s launch material.
The system is being integrated into the Falcon platform, while customers can also pair other models with CrowdStrike’s harnesses. CrowdStrike separately introduced Falcon IQ, an agentic workload-automation system using more than 50 agents to support assessment, prioritisation and remediation workflows.
The launch has a direct operational connection to CrowdStrike’s expansion in India. Deloitte India recently extended its partnership with the company to provide Falcon-based managed security and consolidation services to Indian organisations. CrowdStrike has also announced an in-country Indian cloud deployment intended to provide local data residency while retaining access to its global threat intelligence.
SafeMind is currently offered through an early-access process. CrowdStrike has not announced general-availability dates, pricing or India-specific deployment terms.
