Nvidia and CrowdStrike reckon cyber defenders need AI weapons as quick as the attack bots now snuffling for bugs.
According to WSJ CIO Journal, the pair have introduced SafeMind, a family of agentic AI models designed to find attack paths and shut them down for customers.
Frontier AI models are now spotting software flaws at machine speed, leaving security teams wondering when autonomous hacking models go rogue or start attacking companies.
CrowdStrike and Nvidia say defenders are not getting enough of the same AI firepower. SafeMind is available through CrowdStrike’s Falcon platform and its API, giving customers access to models built for both attack simulation and defence.
CrowdStrike chief AI and autonomous systems officer Bartley Richardson said: “There are all these offensive capabilities of models and systems. CrowdStrike’s point of view is, we should be giving every defender equal and greater capability than every attacker,”.
SafeMind has two models. Red Tempest is the offensive model, built to imitate AI adversaries, while Blue Solano is the defensive model, designed to fix whatever Red Tempest digs up.
The two models are linked by software called a harness, which runs them in a closed-loop system and makes them scrap for improvement. “They co-evolve and learn from each other,” Richardson said.
CrowdStrike built SafeMind by fine-tuning Nvidia’s open Nemotron models with its own threat intelligence and cyber data. Nvidia then tested the system in its own shop and gave CrowdStrike feedback, which is a neat way to make the lab smell like production.
Nvidia CEO Jensen Huang said onstage with CrowdStrike chief executive George Kurtz at the security outfit’s conference, saying: “We’re at an inflexion point in cybersecurity. Instead of what everybody talks about, which is using AI to exploit companies, we’re going to use AI to defend companies.”
Nvidia vice president of enterprise AI, Justin Boitano, said the chipmaker tested SafeMind in a digital twin of its internal technology environment. Red team agents identified attack paths, then Blue team agents applied safeguards until those routes were blocked.
“Being able to get remediation as fast as possible is critical in this conversation,” Boitano said.
CrowdStrike picked Nvidia’s open Nemotron models because it wanted to keep proprietary and customer data private, Richardson said. That matters in cyber, where leaked defensive data can quickly become an attacker’s shopping list.
The fine-tuning of Nvidia’s Nemotron 3 Super model took roughly 45 days from discovery to creation and used about 71 Nvidia B200 GPUs. Other outfits, including AT&T, have been using open-weight and open-source models to cut token costs and keep corporate data closer to home.
Nvidia has led a coalition called the Open Secure AI Alliance, which advocates for open-source AI tools companies can use to defend against cyberattacks. Huang said cyber defence needs open models because specialist jobs need custom kit, not a one-size-fits-all corporate oracle.
“We don’t need every AI to be super smart at everything. But in some areas we need to be extraordinarily good at something, and cyber defence is something we want to be incredibly good at,” Huang said.
