Research map / Adversarial robustness and security / Black-box and transfer attacks
HogVul: Black-box Adversarial Code Generation Framework Against LM-based Vulnerability Detectors
Jingxiao Yang, Ping He, Tianyu Du, Sun Bing, Xuhong Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| combine lexical and syntax perturbations in HogVul versus ALERT (lexical only) | HogVul · Devign, DiverseVul, BigVul (3 victim models) | ASR: +26.37 ASR on average | Improved |
| combine lexical and syntax perturbations in HogVul versus DIP (dead code only) | HogVul · Devign, DiverseVul, BigVul (3 victim models) | ASR: +25.72 ASR on average | Improved |
| HogVul versus ALERT on confidence drop | HogVul · Devign, DiverseVul, BigVul | confidence drop: 0.18 larger than ALERT | Improved |
| HogVul versus DIP on CodeBLEU similarity to the original code | HogVul · Devign, DiverseVul, BigVul | CodeBLEU: Devign CodeT5: 0.86 (DIP) vs 0.84 (HogVul) | Hurt |
| add lexical perturbation (PSO) to structure-only attack (ablation) | HogVul · Devign (CodeT5) | ASR | Improved |
| add structure-aware transformation to lexical-only attack (ablation) | HogVul · Devign (CodeT5) | ASR and confidence drop | Improved |
| combine any two modules (C4-C6) (ablation) | HogVul · Devign (CodeT5) | ASR | Improved |
| full model with semantic-aware initialization, PSO lexical and structural transformation (ablation) | HogVul · Devign (CodeT5) | ASR: best ASR 97.28% | Improved |
| use PSO instead of MHM, greedy search or GA as optimizer (ablation) | HogVul · code search trajectories | convergence and population diversity | Improved |
| HogVul query efficiency versus baselines | HogVul · 9 victim model and dataset scenarios | ASR per Query: highest APQ in six out of nine scenarios | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ALERT — ALERT: Adversarial Learning Enhanced Stability-aware Routing Transformer for Adaptive Depression Detection (AAAI 2026) · better on 11 of 11 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Improved Generation of Adversarial Examples Against Safety-aligned LLMs (NeurIPS 2024 · similarity 0.86)
- Catastrophic Jailbreak of Open-source LLMs via Exploiting Generation (ICLR 2024 · similarity 0.85)
- Black-Box Adversarial Attacks on LLM-Based Code Completion (ICML 2025 · similarity 0.85)
- Understanding and Enhancing the Transferability of Jailbreaking Attacks (ICLR 2025 · similarity 0.85)
- Statistical Estimation of Adversarial Risk in Large Language Models under Best-of-N Sampling (NeurIPS 2026 · similarity 0.85)
- VERA: Variational Inference Framework for Jailbreaking Large Language Models (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
