mlsanityData to 7 Oct 2026

LLM4Vuln: A Unified Evaluation Framework for Decoupling and Enhancing LLMs' Vulnerability Reasoning

arXiv 2401.16185 ↗First posted 29 Jan 2024cs.CR cs.AI cs.SEcited by 57

On the map: LLM reasoning → Reasoning for agents, planning and code

Abstract (arXiv)

Large language models (LLMs) have demonstrated significant potential in various tasks, including those requiring human-level intelligence, such as vulnerability detection. However, recent efforts to use LLMs for vulnerability detection remain preliminary, as they lack a deep understanding of whether a subject LLM's vulnerability reasoning capability stems from the model itself or from external aids such as knowledge retrieval and tooling support. In this paper, we aim to decouple LLMs' vulnerability reasoning from other capabilities, such as vulnerability knowledge adoption, context information retrieval, and advanced prompt schemes. We introduce LLM4Vuln, a unified evaluation framework that separates and assesses LLMs' vulnerability reasoning capabilities and examines improvements when combined with other enhancements. To support this evaluation, we construct UniVul, the first benchmark that provides retrievable knowledge and context-supplementable code across three representative programming languages: Solidity, Java, and C/C++. Using LLM4Vuln and UniVul, we test six representative LLMs (GPT-4.1, Phi-3, Llama-3, o4-mini, DeepSeek-R1, and QwQ-32B) for 147 ground-truth vulnerabilities and 147 non-vulnerable cases in 3,528 controlled scenarios. Our findings reveal the varying impacts of knowledge enhancement, context supplementation, and prompt schemes. We also identify 14 zero-day vulnerabilities in four pilot bug bounty programs, resulting in 3,576 in bounties.

Key facts. Cited by 57 papers in the map (data through 7 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 40 of its 84 references that are in the map. First posted on arXiv 29 Jan 2024.

Builds on

40 of 84 references.

References · 40

  1. 2017Attention Is All You Needcited by 59,932
  2. 2022Training language models to follow instructions with human feedbackcited by 15,712
  3. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  4. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  5. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,160
  6. 2022Large Language Models are Zero-Shot Reasonerscited by 4,323
  7. 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,072
  8. 2017Billion-scale similarity search with GPUscited by 2,477
  9. 2023Code Llama: Open Foundation Models for Codecited by 2,105
  10. 2023Can Large Language Models Be an Alternative to Human Evaluations?cited by 729
  11. 2019Devign: Effective Vulnerability Identification by Learning Comprehensive Program Semantics via Graph Neural Networkscited by 266
  12. 2022Why Can GPT Learn In-Context? Language Models Implicitly Perform Gradient Descent as Meta-Optimizerscited by 204
28 more
  1. 2020Deep Learning based Vulnerability Detection: Are We There Yet?cited by 142
  2. 2021Examining Zero-Shot Vulnerability Repair with Large Language Modelscited by 105
  3. 2022Large Language Models are Zero-Shot Fuzzers: Fuzzing Deep-Learning Libraries via Large Language Modelscited by 102
  4. 2023DiverseVul: A New Vulnerable Source Code Dataset for Deep Learning Based Vulnerability Detectioncited by 98
  5. 2023LLMs Cannot Reliably Identify and Reason About Security Vulnerabilities (Yet?): A Comprehensive Evaluation, Framework, and Benchmarkscited by 94
  6. 2023Fuzz4All: Universal Fuzzing with Large Language Modelscited by 88
  7. 2023Split and Merge: Aligning Position Biases in LLM-based Evaluatorscited by 71
  8. 2017Neural Network-based Graph Embedding for Cross-Platform Binary Code Similarity Detectioncited by 65
  9. 2023GPTScan: Detecting Logic Vulnerabilities in Smart Contracts by Combining GPT with Program Analysiscited by 65
  10. 2023Understanding the Effectiveness of Large Language Models in Detecting Security Vulnerabilitiescited by 58
  11. 2021PalmTree: Learning an Assembly Language Model for Instruction Embeddingcited by 49
  12. 2023Large Language Model-Powered Smart Contract Vulnerability Detection: New Perspectivescited by 48
  13. 2023When ChatGPT Meets Smart Contract Vulnerability Detection: How Far Are We?cited by 42
  14. 2022Transformer-Based Language Models for Software Vulnerability Detectioncited by 39
  15. 2019DIRE: A Neural Approach to Decompiled Identifier Namingcited by 36
  16. 2023Do you still need a manual smart contract audit?cited by 34
  17. 2022SoK: Decentralized Finance (DeFi) Attackscited by 28
  18. 2023TRACED: Execution-aware Pre-training for Source Codecited by 27
  19. 2023How Far Have We Gone in Vulnerability Detection Using Large Language Modelscited by 27
  20. 2021Augmenting Decompiler Output with Learned Variable Names and Typescited by 23
  21. 2018NEUZZ: Efficient Fuzzing with Neural Program Smoothingcited by 19
  22. 2023netFound: Principled Design for Network Foundation Modelscited by 17
  23. 2023Blockchain Large Language Modelscited by 16
  24. 2023Nova: Generative Language Models for Assembly Code with Hierarchical Attention and Contrastive Learningcited by 12
  25. 2023kTrans: Knowledge-Aware Transformer for Binary Code Embeddingcited by 10
  26. 2023Exploiting Code Symmetries for Learning Program Semanticscited by 9
  27. 2022BlockScope: Detecting and Investigating Propagated Vulnerabilities in Forked Blockchain Projectscited by 4
  28. 2020When Program Analysis Meets Bytecode Search: Targeted and Efficient Inter-procedural Analysis of Modern Android Apps in BackDroidcited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 17; 2025: 22; 2026: 18.

57 papers.

Sort:

By area:

  1. 2024Large Language Model-Based Agents for Software Engineering: A Surveycited by 173
  2. 2024Vulnerability Detection with Code Language Models: How Far Are We?cited by 133
  3. 2024Making Them Ask and Answer: Jailbreaking Large Language Models in Few Queries via Disguise and Reconstructioncited by 95
  4. 2024CyberSecEval 2: A Wide-Ranging Cybersecurity Evaluation Suite for Large Language Modelscited by 84
  5. 2024When LLMs Meet Cybersecurity: A Systematic Literature Reviewcited by 70
  6. 2024Vul-RAG: Enhancing LLM-based Vulnerability Detection via Knowledge-level RAGcited by 67
  7. 2024SelfDefend: LLMs Can Defend Themselves against Jailbreaking in a Practical Mannercited by 57
  8. 2024Combining Fine-Tuning and LLM-based Agents for Intuitive Smart Contract Auditing with Justificationscited by 41
  9. 2024PropertyGPT: LLM-driven Formal Verification of Smart Contracts through Retrieval-Augmented Property Generationcited by 36
  10. 2025Benchmarking LLMs and LLM-based Agents in Practical Vulnerability Detection for Code Repositoriescited by 29
  11. 2024Testing and Understanding Erroneous Planning in LLM Agents through Synthesized User Inputscited by 22
  12. 2025Frontier AI's Impact on the Cybersecurity Landscapecited by 21
38 more
  1. 2025Let the Trial Begin: A Mock-Court Approach to Vulnerability Detection using LLM-Based Agentscited by 20
  2. 2025LAMD: Context-driven Android Malware Detection and Classification with LLMscited by 18
  3. 2024Harnessing Large Language Models for Software Vulnerability Detection: A Comprehensive Benchmarking Studycited by 17
  4. 2025Measuring and Augmenting Large Language Models for Solving Capture-the-Flag Challengescited by 14
  5. 2025VulnLLM-R: Specialized Reasoning LLM with Agent Scaffold for Vulnerability Detectioncited by 11
  6. 2024LLMs Can Defend Themselves Against Jailbreaking in a Practical Manner: A Vision Papercited by 9
  7. 2024LProtector: An LLM-driven Vulnerability Detection Systemcited by 8
  8. 2025R2Vul: Learning to Reason about Software Vulnerabilities with Reinforcement Learning and Structured Reasoning Distillationcited by 7
  9. 2025Agent That Debugs: Dynamic State-Guided Vulnerability Repaircited by 7
  10. 2025Smart-LLaMA-DPO: Reinforced Large Language Model for Explainable Smart Contract Vulnerability Detectioncited by 7
  11. 2025Can LLMs Handle WebShell Detection? Overcoming Detection Challenges with Behavioral Function-Aware Frameworkcited by 5
  12. 2025Detecting Various DeFi Price Manipulations with LLM Reasoningcited by 4
  13. 2025A Mixture of Linear Corrections Generates Secure Codecited by 4
  14. 2025LLM Agents for Automated Web Vulnerability Reproduction: Are We There Yet?cited by 4
  15. 2025VulInstruct: Teaching LLMs Root-Cause Reasoning for Vulnerability Detection via Security Specificationscited by 4
  16. 2026From SFT to RL: Demystifying the Post-Training Pipeline for LLM-based Vulnerability Detectioncited by 4
  17. 2024RealVul: Can We Detect Vulnerabilities in Web Applications with LLM?cited by 3
  18. 2024SmartLLMSentry: A Comprehensive LLM Based Smart Contract Vulnerability Detection Frameworkcited by 3
  19. 2025Revisiting Pre-trained Language Models for Vulnerability Detectioncited by 3
  20. 2025Counterfactual Evaluation for Blind Attack Detection in LLM-based Evaluation Systemscited by 3
  21. 2026FuzzingBrain V2: A Multi-Agent LLM System for Automated Vulnerability Discovery and Reproductioncited by 3
  22. 2026DREA: Decoupled Reasoning and Exploration Agents for Repository-Level Vulnerability Detectioncited by 3
  23. 2024Generative transformations and patterns in LLM-native approaches for software verification and falsificationcited by 2
  24. 2025Adaptive Plan-Execute Framework for Smart Contract Security Auditingcited by 2
  25. 2026Revisiting Vul-RAG: Reproducibility and Replicability of RAG-based Vulnerability Detection with Open-Weight Modelscited by 2
  26. 2024Ollabench: Evaluating LLMs' Reasoning for Human-centric Interdependent Cybersecuritycited by 1
  27. 2025Generative AI for Internet of Things Security: Challenges and Opportunitiescited by 1
  28. 2025Experiences with Content Development and Assessment Design in the Era of GenAIcited by 1
  29. 2026A Dual-Loop Agent Framework for Automated Vulnerability Reproductioncited by 1
  30. 2026AEGIS: From Clues to Verdicts -- Graph-Guided Deep Vulnerability Reasoning via Dialectics and Meta-Auditingcited by 1
  31. 2026Comprehensive AI governance requires addressing non-model gainscited by 1
  32. 2026Calibration Without Comprehension: Diagnosing the Limits of Fine-Tuning LLMs for Vulnerability Detection in Systems Softwarecited by 1
  33. 2025LLaVul: A Multimodal LLM for Interpretable Vulnerability Reasoning about Source Codecited by 0
  34. 2025C-ing Clearly: Enhanced Binary Code Explanations using C codecited by 0
  35. 2025SoK: Understanding (New) Security Issues Across AI4Code Use Casescited by 0
  36. 2026VulReaD: Knowledge-Graph-guided Software Vulnerability Reasoning and Detectioncited by 0
  37. 2026QASecClaw: A Multi-Agent LLM Approach for False Positive Reduction in Static Application Security Testingcited by 0
  38. 2026Adversarial SQL Injection Generation with LLM-Based Architecturescited by 0

50 most cited of 57 All 57 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024To Err is Machine: Vulnerability Detection Challenges LLM Reasoning
  2. 2024Towards Explainable Vulnerability Detection with Large Language Models
  3. 2025Boosting Vulnerability Detection of LLMs via Curriculum Preference Optimization with Synthetic Reasoning Data
  4. 2025SV-TrustEval-C: Evaluating Structure and Semantic Reasoning in Large Language Models for Source Code Vulnerability Analysis
  5. 2026Representation Matters: An Empirical Study of Program Representations for LLM Vulnerability Reasoning
  6. 2026Do Fine-Tuned LLMs Understand Vulnerabilities? An Investigation into the Semantic Trap
  7. 2026VulTriage: Triple-Path Context Augmentation for LLM-Based Vulnerability Detection
  8. 2025Reasoning with LLMs for Zero-Shot Vulnerability Detection