Breaking

CrowdStrike and NVIDIA Launch SafeMind With Red-Team and Blue-Team AI Models

September 1, 2026 at 19:14 EDT

AI Agents
Security
Foundation Models

CrowdStrike and NVIDIA have launched SafeMind, an agentic cybersecurity system that pairs two purpose-built AI models—Red Tempest for offensive attack-path discovery and Blue Solano for defensive detection—with a closed-loop harness. The system is built on NVIDIA Nemotron open models and is designed to test attacks against a digital twin of NVIDIA’s accelerated-computing infrastructure. CrowdStrike’s announcement

SEPTEMBER 1, 2026 · CROWDSTRIKE × NVIDIA

Two AI teams, one cyber defense loop

SafeMind pairs Red Tempest and Blue Solano in a continuous cycle that finds attack paths, builds detections and retests the environment.

DETECTION RATE

29%

higher in company evaluations

REMEDIATION

faster end to end

COST SAVINGS

99%

reported reduction

The optimized pipeline lifted backtest detection

Mean detection of recorded attacks; stacked height follows the reported 2.5x improvement.

16.5%

Nemotron 3 Ultra with default harness

41.9%

Optimized open pipeline

How the loop works

RED TEMPEST

Finds an attack path

→

BLUE SOLANO

Generates a detection

→

VALIDATION

Tests and promotes the rule

→

RETEST

The attacker adapts

REPORTED ADVANTAGE

Specialized models with operational tools

The system combines domain data, model specialization and validation harnesses so the agents can act on detected risk.

PUBLISHED CAVEATS

Broader validation is still needed

The study covered one scenario family, limited benign traffic and retained runs with harness failures, so production-scale generalization remains untested.

Continue reading

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