A major American company has unveiled a new information security system. The product was developed jointly with a well-known graphics accelerator manufacturer. The SafeMind system combines offensive and defensive artificial intelligence models. The announcement took place on September 1 at a specialized industry conference. The new product will run natively within the Falcon platform. The system includes two specialized models for a closed loop. The Red Tempest model simulates the actions of a potential network threat actor. The Blue Solano model creates rules for detecting attacks. The software loops provide neural networks with the necessary context and tools.
The defensive loop is built on Nvidia's open Nemotron models. The neural networks have been additionally trained on the company's internal cybersecurity data. The Nemotron 3 Ultra model handles orchestration of defensive agents. The Nemotron 3 Super model is used for writing rules. The training dataset included thousands of error-correction examples. Developers applied additional fine-tuning on cybersecurity materials. The architecture leverages platform telemetry and proprietary threat intelligence. Open weights allow neural networks to be trained without transferring data. CoreWeave's cloud infrastructure participates in training and inference.
Internal evaluations showed a significant improvement in threat detection rates. The new system demonstrated a twenty-nine percent better result. The speed of issue remediation increased sixfold compared to analogues. Detection and remediation costs decreased by ninety-nine percent. Backtesting in an isolated environment confirmed the effectiveness of the specialized loop. The standard model detected a recorded attack in sixteen percent of cases. Adding domain-specific context raised the figure to forty-two percent. The improvement applies to the entire optimized defensive software loop.
Field trials confirmed the high effectiveness of the detection rules created. The open-loop rules successfully identified new, complex network attacks. Three detectors received the highest category and covered all attacks. The compared leading system did not demonstrate similarly outstanding results. The company also unveiled a research division called the Cyber Superintelligence Lab. The lab brings together specialists in artificial intelligence and security. The division was headed by Bartley Richardson in June of this year. Nvidia plans to invest one hundred million dollars over five years. The company announced the Falcon IQ system for task automation. Falcon Guardian protection has been expanded to operate on endpoint devices. In August, the companies signed a letter on strengthening cyber defense.
Experts warn of an imminent increase in the sophistication of mass attacks using neural networks. The open letter was signed by more than one hundred and twenty-seven major global organizations. Developers recognize the urgent need to strengthen global network defenses. New tools will help companies effectively counter modern cybersecurity threats. The information security industry is confidently transitioning toward autonomous defensive systems.