CrowdStrike released a pair of artificial-intelligence models it said were purpose-built for cybersecurity defenses.

The company on Tuesday announced SafeMind, a family of models and model harnesses built in partnership with Nvidia using the chip maker's Nemotron open models. At the core of SafeMind are two models: a model designed to emulate AI cyberattacks called Red Tempest, and a defensive model called Blue Solano built to deploy cyber defenses.

The SafeMind harnesses pit both models against each other to identify and patch vulnerabilities. CrowdStrike said Tuesday that SafeMind had a 29% higher detection rate and a sixfold increase in the speed of end-to-end remediation compared with frontier models and open-source baselines.

The models were trained on CrowdStrike's proprietary data, the company said. It said it used CoreWeave's AI cloud for the models' training and inference.

"With the models and harnesses together in a co-evolving agentic system, defenders can now act at machine speed," Chief AI and Autonomous Systems Officer Bartley Richardson said. "This is the foundation for the next decade of AI security, and CrowdStrike is the only company that owns the entire stack, from sensor to harness to model."

The company on Tuesday also said it was establishing a Cyber Superintelligence Lab to serve as a research hub focused on AI safety and cyberdefense. The lab will be led by Richardson and have access to the company's corpus of internal security data.

CrowdStrike also announced Falcon Guardian, a new AI detection and response solution it said can monitor and protect AI agent activity in enterprises.

Shares in CrowdStrike were down 6.8% to $215.37 in Tuesday afternoon trading. The stock is up 84% this year.