AGENTIC SOLUTION TO SECURE FIREWALL AND CODE VULNERABILITIES

Abstract

Recent Offensive AI Automation in Cyber tools proof the potential of Cyber-attacks which become nightmare for Cyber engineers due to traditional cyber defence. This complexity stems from using disconnected tools, non-standardized data formats, and poor communication among systems for governance, compliance, vulnerability scanning, and network defence. Although Large Language Models (LLMs) recently demonstrated powerful analytical and automated response potential, few current solutions weave these capabilities into a complete, end-to-end framework that can perceive, reason, and safely actuate changes. This paper presents a tiered LLM-led orchestration system that seamlessly combines: industry-standard threat intelligence feeds; a formal safety layer using Agentic reasoning to verify actions; event-driven actuation of firewalls and Access Control Lists (ACLs); and LLM-powered Incident Response (IR) planning with measurable performance metrics. Crucially, the system incorporates long-term memory to build persistent situational awareness and enable adaptive learning. Testing under real-world red-team exercises on threatscan.org company involving coordinated attacks and live network telemetry shows that this orchestrator improves the cyber security engineer’s productivity 10 times through Agentic Ai reasoning and human feedback in detection, response time, remediation and prevention. To the best of our knowledge, this is the first comprehensive architecture to integrate UFW firewall/ACL management, vulnerability assessment, malware containment, and GRC compliance under a formally verified, memory-aware LLM-based planner.

Key words : Agentic AI; LLM 1; ACL 2; Incident Response 3