Skolkovo Resident Develops AI Model to Detect Cyber Threats in Data Streams
Skolkovo resident company NGR Softlab has deployed its proprietary DivergentGPT foundation model into production to detect anomalous user behavior and analyze cybersecurity events.
The model was trained on a combination of real-world and synthetic datasets and serves as a second layer of analysis within the xBA behavioral analytics module of the Dataplan platform. Statistical algorithms first identify suspicious events, after which DivergentGPT determines whether the anomaly represents a genuine security threat or simply normal behavioral variation. This approach reduces false positives while easing the workload of security analysts.
The model is compact and does not require dedicated AI infrastructure. It contains approximately 50 million parameters and occupies about 30 MB of storage.
According to Vladimirov, the model draws on a large body of accumulated experience to help determine whether an anomaly is associated with a genuine security risk.
