Large Language Models in Cybersecurity

Abstract

Large Language Models (LLMs) serve as general-purpose interfaces across security operations,threat intelligence,vulnerability & malware analysis,secure coding,and penetration testing by summarizing evidence & aiding analyst reasoning.Despite rapid adoption,challenges persist in data quality,domain adaptation,privacy,reliability & evaluation. Furthermore,LLMs introduce system-level risks: generating hallucinated content,accepting attacker instructions,exposing sensitive context,retrieving poisoned data,producing insecure code,or abusing tool authority.To address these vulnerabilities,this paper develops a framework structured around explainability, calibrated trust & responsible deployment.It establishes LLM foundations,proposes a six-layer threat taxonomy & evaluates explainability methods.Trust is framed as appropriate reliance through uncertainty calibration & human oversight.Finally,a lifecycle deployment framework integrates data governance,secure architecture,red teaming & regulatory standards(NIST AI RMF,ISO/IEC 42001,EUAI Act)to ensure enforceable,trustworthy security systems.