Research map / Adversarial robustness and security / Attacks on text and graphs
Reranker Helps, but Not Enough: Towards Strong Poisoning Attacks Against Retrieval-Augmented Generation
Xiaokun Yang, Jian Liang, Yesheng Liu, Xin Xiong, Ran He, Tieniu Tan
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 |
|---|---|---|---|
| add a reranker to vanilla RAG | PoisonedRAG-B · NQ, Llama3-8B | ASR (%): 72.6 -> 34.8 (MiniLM) | Hurt |
| add a reranker to vanilla RAG | CorruptRAG-AS · NQ, Llama3-8B | ASR (%): 85.8 -> 3.0 | Hurt |
| add a reranker to vanilla RAG | Prompt Injection · NQ, Llama3-8B | ASR (%): 67.2 -> 2.4 | Hurt |
| add a reranker to vanilla RAG | benign RAG · NQ, Llama3-8B | Acc (%): 33.4 -> 37.4 | Improved |
| add character-level perturbation phase (P3A over P2A) (ablation) | P2A · NQ, MiniLM, Llama3-8B | ASR (%): 63.0 -> 72.2 | Improved |
| add character-level perturbation phase (P3A over P2A) (ablation) | P2A · NQ, MiniLM, Qwen2.5-7B | ASR (%): 67.8 -> 83.4 | Improved |
| add character-level perturbation phase (P3A over P2A) (ablation) | P2A · NQ, vanilla RAG, Llama3-8B | ASR (%): 91.8 -> 82.2 | Hurt |
| add character-level perturbation phase (P3A over P2A) (ablation) | P2A · MS-MARCO, MiniLM, Llama3-8B | ASR (%): 40.6 -> 53.0 | Improved |
| add paraphrasing defense | P3A · NQ, MiniLM, Llama3-8B | ASR (%): 72.2% to 63.8% | Hurt |
| add RobustRAG defense | P3A · NQ, MiniLM, Llama3-8B | ASR (%): 72.2 -> 69.2 | Hurt |
| add TrustRAG defense | P3A · NQ, MiniLM, Llama3-8B | ASR (%): 72.2 -> 48.0 | Hurt |
| transfer attack from MiniLM to ELECTRA reranker | P3A · NQ, Llama3-8B | ASR (%): 72.2 -> 64.4 | Hurt |
Closest papers by meaning
- On the Vulnerability of Applying Retrieval-Augmented Generation within Knowledge-Intensive Application Domains (ICML 2025 · similarity 0.90)
- Cannot See the Forest for the Trees: Invoking Heuristics and Biases to Elicit Irrational Choices of LLMs (ICML 2025 · similarity 0.87)
- PoisonedEye: Knowledge Poisoning Attack on Retrieval-Augmented Generation based Large Vision-Language Models (ICML 2025 · similarity 0.87)
- Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samples (NeurIPS 2026 · similarity 0.86)
- Follow My Instruction and Spill the Beans: Scalable Data Extraction from Retrieval-Augmented Generation Systems (ICLR 2025 · similarity 0.86)
- ReliabilityRAG: Effective and Provably Robust Defense for RAG-based Web-Search (NeurIPS 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.
