Security company CrowdStrike unveiled an autonomous red team system called Safemind that combines offensive AI and defensive AI.
On Sept. 1 local time, SiliconANGLE reported that CrowdStrike, at its annual Fal.Con 2026 conference in Las Vegas, presented a structure in which AI repeatedly searches for attack paths and fixes gaps. The company offered it as an alternative based on its assessment that “breakout time” — the time from intrusion to internal spread — has effectively approached zero.
Safemind consists of two models: the offensive Red Tempest and the defensive Blue Solano. CrowdStrike Cyber Superintelligence Lab developed the two models based on Nvidia Nemotron models.
The two models operate in a closed-loop iterative structure. Red Tempest trained on some of CrowdStrike’s 15 years of incident response data and searches for attack paths by exploring a digital twin of the customer environment within Nvidia simulation technology. Blue Solano remediates confirmed weak points. The process repeats until no paths remain.
Safemind runs natively on the Falcon platform. It can also be accessed separately through the Project Quiltworks program.
Chi-gyu Hwang (황치규)
delight@d-today.co.kr