PulseAugur
EN
LIVE 10:00:14

LLMs susceptible to cognitive biases in code vulnerability detection

By PulseAugur Editorial · [2 sources] ·

A new research paper explores how cognitive heuristics, similar to those affecting human judgment, can influence Large Language Models (LLMs) in detecting code vulnerabilities. The study found that LLMs are susceptible to the halo, framing, and anchoring effects, with framing being the most impactful at 33.2%. This susceptibility can lead models to incorrectly flag code as vulnerable or safe, and researchers demonstrated a black-box attack that could suppress up to 97% of detected vulnerabilities, highlighting a significant exploitable property in LLM-based security tools. AI

IMPACT Reveals exploitable biases in LLM security tools, potentially impacting the reliability of AI-driven code analysis.

RANK_REASON The cluster contains a research paper detailing findings on LLM behavior.

Read on arXiv cs.AI →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

LLMs susceptible to cognitive biases in code vulnerability detection

COVERAGE [2]

  1. arXiv cs.AI TIER_1 English(EN) · Asif Shahriar, Hongyu Cai, Hadjer Benkraouda, Gang Wang, Z. Berkay Celik ·

    Words Speak Louder Than Code: Investigating Cognitive Heuristics in LLM-Based Code Vulnerability Detection

    arXiv:2606.30587v1 Announce Type: cross Abstract: Researchers and practitioners increasingly apply Large Language Models (LLMs) for automated vulnerability detection. Recent work has shown that LLMs are susceptible to the same cognitive heuristics that bias human judgment. Yet, n…

  2. arXiv cs.AI TIER_1 English(EN) · Z. Berkay Celik ·

    Words Speak Louder Than Code: Investigating Cognitive Heuristics in LLM-Based Code Vulnerability Detection

    Researchers and practitioners increasingly apply Large Language Models (LLMs) for automated vulnerability detection. Recent work has shown that LLMs are susceptible to the same cognitive heuristics that bias human judgment. Yet, no work has investigated whether these heuristics a…

RELATED ENTITIES

RELATED TOPICS