Research map / Large language models / Retrieval and knowledge
Silent Leaks: Implicit Knowledge Extraction Attack on RAG Systems
Yuhao Wang, Wenjie Qu, Shengfang Zhai, Yanze Jiang, Liu Zichen, Yue Liu, Yinpeng Dong, Jiaheng 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 |
|---|---|---|---|
| add anchor concepts to random benign queries (ablation) | IKEA · HealthCareMagic | EE: 0.45 -> 0.73 | Improved |
| add experience reflection (ER) (ablation) | IKEA · HealthCareMagic | EE: 0.87 -> 0.92 | Improved |
| add trust region directed mutation (TRDM) (ablation) | IKEA · HealthCareMagic | EE: 0.88 -> 0.92 | Improved |
| add ER and TRDM together (ablation) | IKEA · HealthCareMagic | ASR: 0.90 -> 0.94 | Improved |
| add anchor concepts to random queries (ablation) | IKEA · HealthCareMagic | ASR: 0.97 -> 0.90 | Hurt |
| replace 30% of anchors with related alternatives (ablation) | IKEA · HealthCareMagic | EE: 0.88 -> 0.83 | No clear effect |
| replace 30% of anchors with related alternatives (ablation) | IKEA · HarryPotter | EE: 0.65 -> 0.63 | No clear effect |
| use IKEA instead of RAG-Thief under input-ensemble defense | RAG extraction · HealthCareMagic | EE: 0 -> 0.88 | Improved |
| use IKEA instead of DGEA under no defense | RAG extraction · HealthCareMagic | EE: 0.41 -> 0.87 | Improved |
| use IKEA instead of RAG-thief to build substitute RAG | Pokemon MCQ · Pokemon | accuracy: 0 -> 0.43 | Improved |
| use stricter output Rouge-L content detection | IKEA · HealthCareMagic | EE: 0.87 -> 0.85 | No clear effect |
| use higher trust-region gamma (tighter region) (ablation) | IKEA · three datasets | EE and ASR | Mixed |
Closest papers by meaning
- Follow My Instruction and Spill the Beans: Scalable Data Extraction from Retrieval-Augmented Generation Systems (ICLR 2025 · similarity 0.91)
- SeCon-RAG: A Two-Stage Semantic Filtering and Conflict-Free Framework for Trustworthy RAG (NeurIPS 2025 · similarity 0.90)
- On the Vulnerability of Applying Retrieval-Augmented Generation within Knowledge-Intensive Application Domains (ICML 2025 · similarity 0.90)
- RAGGED: Towards Informed Design of Scalable and Stable RAG Systems (ICML 2025 · similarity 0.88)
- A Theory for Token-Level Harmonization in Retrieval-Augmented Generation (ICLR 2025 · similarity 0.87)
- Speculative RAG: Enhancing Retrieval Augmented Generation through Drafting (ICLR 2025 · similarity 0.87)
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.
