Models

CGN: Cogensec Generative Network

A family of security-focused generative models trained for vulnerability analysis, threat reasoning, and adversarial evaluation. CGN powers Cogensec research and product agents, and is released openly for the security community.

CGN-Reason

Specialized for long-horizon security reasoning across vulnerability reports, threat intelligence, and incident-response playbooks.

Training data: Security Reasoning Corpus

CGN-Attack

An adversarial evaluation model that generates and refines jailbreaks, prompt injections, and multi-turn attack chains for red teaming.

Training data: Jailbreak Corpus

Architecture principles

  • Mixture-of-experts routing for security-specific tasks
  • Constitutional guardrails aligned to NIST and OWASP frameworks
  • Tool-use grounding with deterministic verification hooks
  • Adversarial training on curated attack corpora
  • Interpretable confidence scoring for risk amplification
  • Deployment-ready quantization for edge and air-gapped environments