본문 바로가기
  • Home

A Cognitive-Aware Security Framework for LLM-Based Social Engineering

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2026, 31(8), pp.63~72
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : April 30, 2026
  • Accepted : July 30, 2026
  • Published : August 31, 2026

Woo Jin Jung 1 Ah Reum Kang 2

1고려대학교
2배재대학교

Accredited

ABSTRACT

The emergence of large language model (LLM)-based generative artificial intelligence is fundamentally transforming social engineering attacks by enabling large-scale automation, extreme personalization, and context-aware persuasion. As a result, such attacks can no longer be explained solely by user carelessness but increasingly exploit inherent vulnerabilities in human cognition. This paper reconceptualizes LLM-based social engineering as a problem of cognitive vulnerability amplification, in which linguistic naturalness, authority cues, contextual dependence, and cognitive overload are systematically leveraged to manipulate decision-making. To address this shift, we propose a cognitive-aware security framework that moves beyond content-based detection and focuses on protecting user decision processes through interaction-level analysis and decision-point intervention. Through structured analysis and a case study, we discuss the limitations of conventional security models and highlight the need for a human-centric defense paradigm in the era of generative AI.

Citation status

* References for papers published after 2025 are currently being built.