NVIDIA CEO Jensen Huang and CrowdStrike CEO George Kurtz announced CrowdStrike SafeMind at CrowdStrike Fal.Con 2026, held in Las Vegas on Tuesday. Speaking at the event, which was attended by 10,000 security professionals, Huang said cyberattacks had become automated and that defense must also be automated. Kurtz said attackers were using frontier AI, while defenders would now have access to the same capability.
Developed by the CrowdStrike Cyber Superintelligence Lab, SafeMind combines NVIDIA Nemotron open models retrained with the company’s cybersecurity expertise and threat data with specialized cybersecurity tools. The system aims to strengthen the security of customer environments by operating in a cycle in which attack and defense models continuously test and improve one another. NVIDIA Nemotron 3 Ultra manages the defense agent, while Nemotron 3 Super powers SafeMind’s rule-generation sub-agent.
According to CrowdStrike’s internal evaluations, the Nemotron 3 Super-based Blue Solano model delivered higher accuracy than leading frontier models while keeping costs 99% lower. The models can be used independently, and customers can pair their own models with CrowdStrike’s tools.
The company also announced CrowdStrike Falcon IQ to automate Project QuiltWorks and expanded the CrowdStrike Guardian AI security solution.
Why it matters
The SafeMind announcement shows that AI in cybersecurity is being positioned not merely as a tool that analyzes threats, but as a system developed by testing attack and defense scenarios against one another. This approach is relevant to customers who want to combine threat data with their own operations and maintain flexibility over the models they use; the ability to run the models independently and pair them with CrowdStrike tools expands this option. While the accuracy and cost difference reported in the company’s internal evaluations provides a basis for broader adoption, it remains unclear how these results will hold up in independent measurements and real customer environments. The Falcon IQ and Guardian AI announcements also complement the approach of automating different security tasks within the same product ecosystem.
Background
NVIDIA is not a new name in the FikirPilot archive: we have published 3 news stories mentioning the name in the last 90 days; the most recent is dated August 31, 2026.
Term: agent
An AI agent is software that calls tools and carries out multi-step tasks to achieve a goal rather than producing a single response.