Breaking

CrowdStrike Unveils SafeMind, Pairing Offensive and Defensive AI Models With NVIDIA

September 6, 2026 at 09:41 EDT

Foundation Models
AI Agents

CrowdStrike has unveiled SafeMind, a family of purpose-built security models and agent harnesses built with NVIDIA that the company describes as the first complete agentic system designed for cyber defenders. The system was introduced on September 1, 2026 at Fal.Con 2026 in Las Vegas, where CrowdStrike CEO and founder George Kurtz appeared with NVIDIA founder and CEO Jensen Huang.

September 1, 2026 · CrowdStrike · NVIDIA

SafeMind puts red and blue agents in one loop

CrowdStrike launched purpose-built security models with NVIDIA: Red Tempest hunts attack paths, Blue Solano closes them, and harnesses keep both running inside Falcon.

29%

Higher detection vs unnamed frontier and open baselines (vendor test)

Faster end-to-end remediation in CrowdStrike evaluations

99%

Claimed cost cut on detection and remediation

$100M

NVIDIA investment in the Cyber Superintelligence Lab over 5 years

Remediation speed, drawn to scale

CrowdStrike says SafeMind finishes detection-to-remediation 6 times faster than the frontier baselines it tested.

Frontier baseline

SafeMind claimed speed

Closed-loop defense

1 · OFFENSE

Red Tempest

Emulates AI adversaries on a Falcon digital twin

→

2 · DEFENSE

Blue Solano

Writes, tests, and deploys detections

→

3 · LOOP

Adapt and repeat

Red retries until no viable path remains

What supporters see

A domain-specific stack that can act on risk, trained on Falcon telemetry rather than a general chatbot with a security skin.

What remains unproven

Headline scores are CrowdStrike internal tests against unnamed models. No general availability or pricing yet; NVIDIA's related eval covered one scenario family.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.