CrowdStrike and NVIDIA Partnership Pushes AI Cybersecurity

The CrowdStrike and NVIDIA partnership is taking a more specialised approach to AI cybersecurity with CrowdStrike SafeMind, a family of security models and harnesses designed specifically for defenders. The system will operate natively within the CrowdStrike Falcon platform, bringing offensive and defensive AI into a continuous loop.

Rather than using general-purpose frontier models alone, SafeMind is built around two distinct models. Red Tempest is designed to simulate advanced attack scenarios, while Blue Solano focuses on defensive measures used to protect enterprise assets. The idea is straightforward: one model finds weaknesses, the other works to close them, and the system learns through repeated cycles.

SafeMind puts cybersecurity data at the centre

A major part of the approach is the data used to train the models. CrowdStrike says SafeMind draws on Falcon sensor telemetry, threat intelligence, Falcon Complete MDR event annotations and 15 years of incident response work.

That gives the system a security-specific training base rather than relying only on broader AI datasets. The models and their supporting harnesses are designed to operate together, continuously testing offensive techniques against defensive responses.

The harnesses can also work with frontier and open-source models. This gives users flexibility in model choice while supporting cost control, according to the company.

NVIDIA brings the AI infrastructure

The CrowdStrike and NVIDIA partnership extends beyond model development. CrowdStrike is building SafeMind using NVIDIA Nemotron open models, with NVIDIA serving as its AI design partner. CoreWeave's AI Cloud will support training and inference.

This combination points to the infrastructure demands behind AI cybersecurity agents. SafeMind is designed to operate continuously and autonomously, making computing capacity part of the system rather than simply a supporting layer.

NVIDIA CEO Jensen Huang described cybersecurity as one of the more compute-intensive AI applications as attackers and defenders increasingly use AI to scale their capabilities.

Early evaluations point to speed and cost

CrowdStrike reports that SafeMind delivered a 29% higher detection rate, six times faster end-to-end remediation and 99% cost savings on detection and remediation compared with leading frontier models and open-source baselines.

These results frame the system around more than detection. The emphasis is on acting on identified risk. CrowdStrike's approach is to connect the model, harness and cybersecurity data so that AI can move from identifying a problem to responding to it.

Read More: 

Amit Chaudhary Appointment Expands Quick Heal Product Mandate

Cloudflare adaptive intelligence changes the bot fight

Lenovo appoints Kiran Balasubramanian to lead India communications

Solitaire Unveils AI-Enabled 4K Interactive Flat Panel Displays for Education and Enterprise