AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Through the Stealth Lens: Attention-Aware Defenses Against Poisoning in RAG

Sarthak Choudhary, Nils Palumbo, Ashish Hooda, Krishnamurthy Dj Dvijotham, Somesh Jha

ICML 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: LLM jailbreak and injection defense · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
add AV Filter to vanilla RAGRAG pipeline · RQA-MC clean (avg over LLMs)clean accuracy: 61.4 -> 59.3Hurt
add AV Filter to vanilla RAGRAG pipeline · RQA-MC under attack (avg)robust accuracy: 44.4 -> 55.7Improved
add AV Filter to vanilla RAGRAG pipeline · RQA-MC under attack (avg)attack success rate: 88.2 -> 6.6Improved
use AV Filter instead of Keyword defenseRAG pipeline · RQA-MC under attack (avg)robust accuracy: 53.9 -> 55.7Improved
use AV Filter instead of Keyword defenseRAG pipeline · RQA-MC under attack (avg)attack success rate: 6.1 -> 6.6No clear effect
use AV Filter instead of Decoding defenseRAG pipeline · RQA-MC clean (avg over LLMs)clean accuracy: 50.3 -> 59.3Improved
combine AV Filter with KeywordKeyword defense · RACC and ASR across settingsrobust accuracy and ASR: 18/18 casesImproved
add adaptive attention-aware attack against AV FilterAV Filter · RQA-MCattack success rate: 6.6 -> 20Hurt
use AV Filter instead of Keyword on multi-hop QARAG pipeline · HotpotQAattack success rateImproved
vary NormScore alpha from 5 to infinity (ablation)AV Filter · RQA-MC clean (Mistral-7B)clean accuracy: 73.0 -> 76.0Improved
vary NormScore alpha from 5 to infinity (ablation)AV Filter · RQA-MC clean (Llama2-C)clean accuracy: 79.0 -> 75.0Hurt
use AV Filter instead of Keyword on clean Llama2-CRAG pipeline · RQA-MC clean (Llama2-C)clean accuracy: 56.0 -> 79.0Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy