NVIDIA and CrowdStrike Unveil SafeMind as AI-Powered Cyberattacks Force a New Defense Era

LAS VEGAS — September 1, 2026 — NVIDIA CEO Jensen Huang and CrowdStrike CEO George Kurtz are warning that cybersecurity has reached a turning point as artificial intelligence allows attackers to operate at machine speed.

Speaking at CrowdStrike’s Fal.Con 2026 in Las Vegas, Huang said the industry is at an “inflection point in cybersecurity,” arguing that automated attacks require automated defenses. The companies used the event to unveil CrowdStrike SafeMind, an agentic cybersecurity system designed to continuously test, adapt and strengthen cyber defenses using artificial intelligence.

The announcement brings NVIDIA’s Nemotron open models together with CrowdStrike’s cybersecurity data, threat intelligence and specialized AI agents. Rather than functioning as a traditional security chatbot or copilot, SafeMind is designed as an offensive-and-defensive system in which AI agents simulate attacks, develop defenses and continuously challenge one another.

“This is the beginning of a new age of cybersecurity,” Huang told an audience of roughly 10,000 security professionals at the sold-out conference.

CrowdStrike SafeMind Brings Red-Team and Blue-Team AI Together

SafeMind is built around a continuous “coevolution” approach to cybersecurity.

CrowdStrike says its defensive models are built using NVIDIA Nemotron open models and post-trained with the company’s cybersecurity experience and threat data. Those models are paired with proprietary cybersecurity “harnesses” that allow them to operate as agents rather than simply generating text or recommendations.

The system includes two specialized model capabilities: Red Tempest, designed for offensive security scenarios, and Blue Solano, designed for defensive operations.

The idea is straightforward: one AI attempts to find an attack path while another works to close it. The results can then be converted into detections and defensive measures.

CrowdStrike says SafeMind operates natively within its Falcon platform, allowing the AI system to work with the company’s broader cybersecurity infrastructure.

Kurtz described the technology as a fundamental shift from conventional AI assistants.

“The real gap that I saw was that the attackers had frontier AI, and the defenders didn’t,” Kurtz said. “And that changes now.”

NVIDIA Nemotron Powers the Defensive AI

NVIDIA’s role goes beyond supplying computing hardware.

The companies are using NVIDIA Nemotron open models as a foundation for CrowdStrike’s cybersecurity-specific AI. CrowdStrike then post-trains those models using its own threat intelligence and cybersecurity experience.

According to CrowdStrike and NVIDIA, Nemotron 3 Ultra orchestrates the defensive agent harness, while a fine-tuned Nemotron 3 Super model powers SafeMind’s rule-generation capabilities.

CrowdStrike says internal evaluations showed its Blue Solano model, based on Nemotron 3 Super, achieved higher accuracy than leading frontier models while costing 99% less. That figure is a company-reported internal evaluation rather than an independently verified industry benchmark.

That distinction is important because cybersecurity organizations increasingly have to balance AI performance with the cost of running models at scale.

For security teams, the companies argue that using an open model as a foundation also provides greater flexibility to customize AI around an organization’s own environment and threat data.

Why AI Is Changing Cybersecurity

The push toward agentic cybersecurity comes as attackers increasingly use automation and AI to accelerate their operations.

CrowdStrike says AI-enabled attacks increased 89% over the past year and that the fastest observed eCrime breakout time has fallen to just 27 seconds.

At those speeds, a defense process that depends entirely on humans manually investigating alerts can struggle to keep pace.

That is the problem SafeMind is designed to address.

Instead of waiting for analysts to identify every new attack technique, AI agents can continuously simulate adversarial behavior, search for weaknesses and generate defensive responses.

Huang compared the relationship between an AI model and its agentic harness to a brain and an exoskeleton.

The model provides the intelligence, while the harness gives that intelligence the ability to operate as an agent within a specific environment.

NVIDIA and CrowdStrike Test SafeMind in a Digital Twin

The companies are also testing SafeMind in a simulated environment modeled on NVIDIA’s own accelerated-computing infrastructure.

In the test environment, a red-team agent attempts to identify and exploit weaknesses while a blue-team agent monitors activity, develops detections and works to close the vulnerabilities.

The offensive system uses specialized Recon, Assault and Compromise sub-agents to execute attack paths. The defensive system uses Falcon sensors to monitor activity, generate detection candidates, validate them and promote successful detections.

The approach effectively creates an AI-powered cybersecurity cat-and-mouse game.

Each side is designed to learn from the other, with the goal of making attacks increasingly difficult to execute successfully.

The companies say the same basic framework could eventually have applications beyond enterprise cybersecurity, including robotics, edge computing and other autonomous systems.

Falcon IQ Adds More AI Agents to Security Workflows

SafeMind was not the only major AI announcement at Fal.Con 2026.

CrowdStrike also introduced Falcon IQ, a system designed to automate cybersecurity assessment, prioritization and remediation workflows.

Falcon IQ uses more than 50 agents as a coordinated AI workforce. NVIDIA Nemotron models help power the agentic engine inside CrowdStrike’s Charlotte AI AgentWorks platform.

The system is designed to help CrowdStrike partners produce customized findings, recommendations and executive-level outputs while reducing the amount of manual work required from security teams.

CrowdStrike said Falcon IQ operationalizes its Project QuiltWorks initiative, which focuses on securing frontier AI risks across the cybersecurity ecosystem.

CrowdStrike Is Building a Broader AI Security Stack

The SafeMind announcement comes as CrowdStrike expands its broader strategy around AI security.

The company also introduced Falcon Guardian, an AI detection and response system designed to provide visibility and runtime enforcement for AI agents operating across enterprise environments. CrowdStrike says Guardian is intended to identify AI activity and help stop threats at runtime, including risks involving AI agents.

CrowdStrike has also announced integrations with Google Cloud’s enterprise AI ecosystem, extending Falcon Guardian through Google Agent Gateway to help address risks such as prompt injection, sensitive data leakage and malicious AI activity.

Together, the announcements show that CrowdStrike is positioning cybersecurity as a full-stack AI problem — from protecting AI models and agents to using AI itself to defend enterprise systems.

Why the NVIDIA-CrowdStrike Partnership Matters

The partnership combines two different advantages.

NVIDIA brings accelerated computing infrastructure and the Nemotron family of open models, while CrowdStrike brings years of cybersecurity telemetry, threat intelligence, incident-response experience and a large installed base of security sensors.

CrowdStrike says its SafeMind models incorporate data from Falcon telemetry, threat intelligence, Falcon Complete MDR event annotations and 15 years of incident-response fieldwork.

That combination could give specialized cybersecurity models an advantage over general-purpose AI systems when dealing with highly technical security tasks.

The bigger question is whether AI defenders can improve quickly enough to keep pace with AI-powered attackers.

That race is likely to become one of the defining battles of enterprise cybersecurity.

What Happens Next

CrowdStrike’s Fal.Con 2026 runs through September 3 at Mandalay Bay in Las Vegas, with more than 10,000 attendees expected at the sold-out event.

For SafeMind, the next major test will be how the technology performs in real-world customer environments rather than controlled simulations and company evaluations.

If the system can reliably turn adversarial activity into automated defensive improvements while maintaining accuracy and controlling costs, it could represent a significant shift in how security operations are conducted.

For now, NVIDIA and CrowdStrike are betting that the future of cybersecurity will not simply involve humans using AI.

It will involve AI systems fighting AI systems — continuously, automatically and at machine speed.