Research map / Large language models / Retrieval and knowledge
Joint-GCG: Unified Gradient-Based Poisoning Attacks on Retrieval-Augmented Generation Systems
Haowei Wang, Rupeng Zhang, Junjie Wang, Mingyang Li, Yuekai Huang, Dandan Wang, Qing Wang
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 |
|---|---|---|---|
| jointly optimise retriever and generator losses (Joint-GCG) instead of GCG | RAG poisoning on Llama3 · NQ (Contriever) | ASR_gen (%): 72.0% -> 92.0% | Improved |
| use Joint-GCG instead of LIAR | RAG poisoning on Llama3 · NQ (Contriever) | ASR_gen (%): 89.0% -> 92.0% | Improved |
| use Joint-GCG instead of LIAR | RAG poisoning on Qwen2 · HotpotQA (Contriever) | ASR_gen (%): 98.0% -> 99.0% | Improved |
| add CVP and GTA gradient alignment (ablation) | Joint-GCG · MS MARCO Llama3 (Contriever) | ASR_gen (%): 93.33% -> 94.00% | Improved |
| add retriever-side loss Lret (ablation) | Joint-GCG · MS MARCO Llama3 (Contriever) | ASR_gen (%): 91.00% -> 94.00% | Improved |
| add retriever-side loss Lret (ablation) | Joint-GCG · NQ Llama3 (Contriever) | ASR_gen (%): 86.67% -> 92.00% | Improved |
| use adaptive weighted fusion instead of fixed weights (ablation) | Joint-GCG · Contriever/Llama3 | ASR (%) | Improved |
| compute AWF stability on synthetic corpus instead of real top-k (ablation) | Joint-GCG · MS MARCO Llama3 (Contriever) | ASR_gen (%): 94.00% -> 62.00% | Hurt |
| optimise poison on surrogate generator then transfer | Joint-GCG · MS MARCO (BGE) | ASR_gen (%): 41% | Improved |
| add Joint-GCG optimisation to poison for black-box GPT-4o | Joint-GCG · GPT-4o transfer | ASR (%): +2% | Improved |
Closest papers by meaning
- On the Vulnerability of Applying Retrieval-Augmented Generation within Knowledge-Intensive Application Domains (ICML 2025 · similarity 0.91)
- Follow My Instruction and Spill the Beans: Scalable Data Extraction from Retrieval-Augmented Generation Systems (ICLR 2025 · similarity 0.89)
- NeuroGenPoisoning: Neuron-Guided Attacks on Retrieval-Augmented Generation of LLM via Genetic Optimization of External Knowledge (NeurIPS 2025 · similarity 0.87)
- RAGGED: Towards Informed Design of Scalable and Stable RAG Systems (ICML 2025 · similarity 0.87)
- PoisonedEye: Knowledge Poisoning Attack on Retrieval-Augmented Generation based Large Vision-Language Models (ICML 2025 · similarity 0.87)
- A Theory for Token-Level Harmonization in Retrieval-Augmented Generation (ICLR 2025 · similarity 0.86)
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.
