Model Releases
Building an Adaptive Agentic Cybersecurity System with NVIDIA Nemotron
NVIDIA and CrowdStrike evaluated an “agentic” cyber‑defense platform that fuses NVIDIA’s Nemotron open‑model family with Falcon telemetry for a continuous offense‑defense closed loop running at machin
NVIDIA and CrowdStrike evaluated an “agentic” cyber‑defense platform that fuses NVIDIA’s Nemotron open‑model family with Falcon telemetry for a continuous offense‑defense closed loop running at machine speed on isolated NVIDIA‑accelerated infrastructure. The system uses a six‑mechanism defensive harness—schema knowledge, telemetry grounding, specialized detection authoring via Nemotron 3 Super, artifact linting, detection replay, and independent review—to generate grounded, testable detections; post‑training with NeMo Gym and NeMo RL improves performance. Backtesting showed a 41.9 % mean detection rate (2.5× better than Nemotron 3 Ultra alone) and live‑fire tests achieved 45 % generalization versus 29 % for the baseline, with three “gold” detections covering all eight unseen attacks.
Related
- NVIDIA Nemotron 3.5 Lightning Delivers Fast, Accurate Specialized Task Execution for Long-Running Agents
- NVIDIA Nemotron 3 Ultra Leads Open Models on Accuracy and Efficiency in Agentic RTL Coding
- NVIDIA Nemotron 3 Ultra Powers Faster, More Efficient Reasoning for Long-Running Agents
- NEW: NVIDIA ships 550B MoE open model for long-running agents. Very exciting times to see more open models to support local long-running cod…
Source: NVIDIA Developer | 2026-09-01