NVIDIA and CrowdStrike Strengthen Agentic Cybersecurity Frontier

CrowdStrike, a cybersecurity company, says its SafeMind system is built around a loop in which simulated attackers and defenders improve each other. NVIDIA's account of the announcement describes a test in a digital twin, a simulated copy, of NVIDIA's own computing infrastructure. Red-team agents attempt attacks there, while blue-team agents monitor activity through CrowdStrike Falcon sensors, generate candidate detection rules and validate them. Findings from successful exploits are meant to become detections that block those attacks.

The defensive models are built on NVIDIA's open Nemotron models, post-trained on CrowdStrike's threat data. CrowdStrike says the loop continuously hardens a customer's environment until attacks are unsuccessful. That is the companies' design goal and account of their own testing; the article gives no independent results and does not establish that SafeMind prevents breaches.