コミュニティで最も包括的かつ継続的に更新される、ソフトウェア脆弱性検出のための大規模言語モデルに関する研究インデックス — 関数レベル、リポジトリレベル、エージェント型、スマートコントラクト検出にわたる論文に加え、データセット、ベンチマーク、サーベイを収録。
# 脆弱性検出のための大規模言語モデル Awesome リスト LLM を脆弱性の検出と発見に活用するための論文、プロジェクト、エージェントスキルの厳選リストです。 --- ## 📄 論文 > 2025年以降のみを表示しています。それ以前の研究については、[論文アーカイブ(2024年以前)](https://github.com/huhusmang/awesome-llms-for-vulnerability-detection/blob/main/docs/papers_archive.md)を参照してください。 | Title | Venue | Year | Paper | Github | | :---: | :---: | :--: | :---: | :---: | | VulnGym: Benchmarking Coding Agents for Repository-Level Vulnerability Detection | | 2026 | [link](https://arxiv.org/abs/2608.02001) | [link](https://github.com/Tencent/VulnGym) | | VulTriage: Triple-Path Context Augmentation for LLM-Based Vulnerability Detection | | 2026 | [link](https://arxiv.org/abs/2605.09461) | [link](https://github.com/vinsontang1/VulTriage) | | Synthesizing Multi-Agent Harnesses for Vulnerability Discovery | | 2026 | [link](https://arxiv.org/abs/2604.20801) | [link](https://github.com/berabuddies/agentflow) | | QRS: A Rule-Synthesizing Neuro-Symbolic Triad for Autonomous Vulnerability Discovery | | 2026 | [link](https://arxiv.org/abs/2602.09774) | | | Seclens: Role-specific Evaluation of LLM's for security vulnerablity detection | | 2026 | [link](https://arxiv.org/abs/2604.01637) | [link](https://github.com/mattersec-labs/seclens) | | Do Fine-Tuned LLMs Understand Vulnerabilities? An Investigation into the Semantic Trap | | 2026 | [link](https://arxiv.org/abs/2601.22655) | | | Sifting the Noise: A Comparative Study of LLM Agents in Vulnerability False Positive Filtering | ISSTA | 2026 | [link](https://arxiv.org/abs/2601.22952) | | | AgenticSCR: An Autonomous Agentic Secure Code Review for Immature Vulnerabilities Detection | | 2026 | [link](https://arxiv.org/abs/2601.19138) | | | LLM-based Vulnerability Detection at Project Scale: An Empirical Study | | 2026 | [link](https://arxiv.org/abs/2601.19239) | | | MulVul: Retrieval-augmented Multi-Agent Code Vulnerability Detection via Cross-Model Prompt Evolution | | 2026 | [link](https://arxiv.org/abs/2601.18847) | | | VulInstruct: Teaching LLMs Root-Cause Reasoning for Vulnerability Detection via Security Specifications | FSE | 2026 | [link](https://arxiv.org/abs/2511.04014) | [link](https://github.com/zhuhaopku/VulInstruct) | | VulnLLM-R: Specialized Reasoning LLM with Agent Scaffold for Vulnerability Detection | | 2025 | [link](https://arxiv.org/abs/2512.07533) | | | VULPO: Context-Aware Vulnerability Detection via On-Policy LLM Optimization | | 2025 | [link](https://arxiv.org/abs/2511.11896) | | | From Large to Mammoth: A Comparative Evaluation of Large Language Models in Vulnerability Detection | NDSS | 2025 | [link](https://www.ndss-symposium.org/ndss-paper/from-large-to-mammoth-a-comparative-evaluation-of-large-language-models-in-vulnerability-detection/) | | | Benchmarking LLMs and LLM-based Agents in Practical Vulnerability Detection for Code Repositories | ACL | 2025 | [link](https://aclanthology.org/2025.acl-long.1490/) | [link](https://github.com/alperen21/JitVul) | | A Systematic Literature Review on Detecting Software Vulnerabilities with Large Language Models | | 2025 | [link](https://arxiv.org/abs/2507.22659) | [link](https://github.com/hs-esslingen-it-security/Awesome-LLM4SVD) | | LLMxCPG: Context-Aware Vulnerability Detection Through Code Property Graph-Guided Large Language Models | Usenix | 2025 | [link](https://arxiv.org/abs/2507.16585) | [link](https://github.com/qcri/llmxcpg) | | CLeVeR: Multi-modal Contrastive Learning for Vulnerability Code Representation | ACL Findings | 2025 | [link](https://aclanthology.org/2025.findings-acl.414/) | [link](https://github.com/yoimiya-nlp/CLeVeR) | | Mono: Is Your "Clean" Vulnerability Dataset Really Solvable? Exposing and Trapping Undecidable Patches and Beyond | | 2025 | [link](https://arxiv.org/abs/2506.03651) | [link](https://github.com/vul337/mono) | | Learning to Focus: Context Extraction for Efficient Code Vulnerability Detection with Language Models | | 2025 | [link](https://arxiv.org/abs/2505.17460) | | | SV-TrustEval-C: Evaluating Structure and Semantic Reasoning in Large Language Models for Source Code Vulnerability Analysis | SP | 2025 | [link](https://arxiv.org/abs/2505.20630) | [link](https://github.com/Jackline97/SV-TrustEval-C) | | SecVulEval: Benchmarking LLMs for Real-World C/C++ Vulnerability Detection | | 2025 | [link](https://arxiv.org/abs/2505.19828) | [link](https://github.com/basimbd/secvuleval) | | CVE-Bench: Benchmarking LLM-based Software Engineering Agent's Ability to Repair Real-World CVE Vulnerabilities | NAACL | 2025 | [link](https://aclanthology.org/2025.naacl-long.212/) | [link](https://github.com/WhileBug/CVEBench) | | R2Vul: Learning to Reason about Software Vulnerabilities with Reinforcement Learning and Structured Reasoning Distillation | | 2025 | [link](https://arxiv.org/abs/2504.04699) | [link](https://github.com/martin-wey/R2Vul) | | Neuro-symbolic Static Analysis with LLM-generated Vulnerability Patterns | | 2025 | [link](https://arxiv.org/abs/2504.16057) | | | Context-Enhanced Vulnerability Detection Based on Large Language Model | | 2025 | [link](https://arxiv.org/abs/2504.16877) | | | Everything You Wanted to Know About LLM-based Vulnerability Detection But Were Afraid to Ask | | 2025 | [link](https://arxiv.org/abs/2504.13474) | [link](https://anonymous.4open.science/r/CORRECT/README.md) | | MOS: Towards Effective Smart Contract Vulnerability Detection through Mixture-of-Experts Tuning of Large Language Models | | 2025 | [link](https://arxiv.org/abs/2504.12234) | | | Abundant Modalities Offer More Nutrients: Multi-Modal-Based Function-Level Vulnerability Detection | TOSEM | 2025 | [link](https://dl.acm.org/doi/10.1145/3731557) | [link](https://github.com/vinci-grape/MVulD) | | Generative Large Language Model usage in Smart Contract Vulnerability Detection | | 2025 | [link](https://arxiv.org/abs/2504.04685) | | | Closing the Gap: A User Study on the Real-world Usefulness of AI-powered Vulnerability Detection & Repair in the IDE | ICSE | 2025 | [link](https://www.arxiv.org/abs/2412.14306) | [link](https://doi.org/10.6084/m9.figshare.26367139) | | Vulnerability Detection with Code Language Models: How Far Are We? | ICSE | 2025 | [link](https://arxiv.org/abs/2403.18624) | [link](https://github.com/DLVulDet/PrimeVul) | | Combining Fine-Tuning and LLM-based Agents for Intuitive Smart Contract Auditing with Justifications | ICSE | 2025 | [link](https://arxiv.org/abs/2403.16073) | | | LAMD: Context-driven Android Malware Detection and Classification with LLMs | | 2025 | [link](http://arxiv.org/abs/2502.13055) | | | LLMs in Software Security: A Survey of Vulnerability Detection Techniques and Insights | | 2025 | [link](https://arxiv.org/abs/2502.07049) | [link](https://github.com/OwenSanzas/LLM-For-Software-Security) | | One-for-All Does Not Work! Enhancing Vulnerability Detection by Mixture-of-Experts (MoE) | | 2025 | [link](https://arxiv.org/abs/2501.16454) | | ## 🚀 プロジェクト | 名前 | 説明 | Github | | :---: | :---: | :---: | | OpenAnt (Knostic) | LLMを活用した多段階の脆弱性発見と敵対的検証 | [link](https://github.com/knostic/OpenAnt) | | DeepAudit | Dockerサンドボックスによるエクスプロイト検証を備えたマルチエージェントAIレッドチームプラットフォーム | [link](https://github.com/lintsinghua/DeepAudit) | | AutoCVE | マルチエージェントアーキテクチャによる自動脆弱性検出とレポート作成 | [link](https://github.com/larlarua/AutoCVE) | | Darkmoon | オープンソース(GPL-3.0)の自律型AIペネトレーションテストプラットフォーム兼MCPホスト。技術ごとの攻撃用サブエージェント、Active DirectoryおよびKubernetesへの対応、80以上のオーケストレーションされたツール、検出項目ごとの証跡 | [link](https://github.com/ASCIT31/Dark-Moon) | | strix | pipで導入できるオープンソースの自律型AIペネトレーションテストツール | [link](https://github.com/usestrix/strix) | | deepsec (Vercel) | コーディングエージェントによる大規模コードベースの脆弱性スキャン向けセキュリティハーネス | [link](https://github.com/vercel-labs/deepsec) | | rust-in-peace | エージェント支援によるRustの脆弱性発見。敵対的トリアージ、動的検証、実験的なAndroid APK解析プロファイルを備える | [link](https://github.com/scadastrangelove/rust-in-peace) | --- ## 🧩 エージェントスキル | 名前 | 説明 | リンク | | :---: | :---: | :---: | | codex-security (OpenAI) | Codexエージェントによるリポジトリ単位の自律型脆弱性スキャン | [link](https://openai.com/index/codex-security-now-in-research-preview/) | | defending-code-reference-harness (Anthropic) | Claude Codeによる脅威モデリング、スキャン、トリアージ、パッチ適用のためのリファレンススキル | [link](https://github.com/anthropics/defending-code-reference-harness) | | security-audit-skill (Cloudflare) | 並列ハンティングエージェントと敵対的検証を備えた6フェーズのセキュリティ監査スキル | [link](https://github.com/cloudflare/security-audit-skill) | --- ## 📡 arxiv.md ワークフローを通じて、指定キーワードに関するArxiv論文を毎日自動で取得・更新します。 ## 謝辞 本プロジェクトのUpdated Arxiv Papers Dailyワークフローは、[LLM4SE](https://github.com/gai4se/LLM4SE)プロジェクトから借用しています。元のコードは[arxiv](https://github.com/lukasschwab/arxiv.py)ライブラリを使用する形にリファクタリングしました。