AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Doxing via the Lens: Revealing Location-related Privacy Leakage on Multi-modal Large Reasoning Models

Weidi Luo, Tianyu Lu, Qiming Zhang, Xiaogeng Liu, Bin Hu, Yue Zhao, Jieyu Zhao, Song Gao, Patrick McDaniel, Zhen Xiang, Chaowei Xiao

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Faithfulness of LLM explanations · 29 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
use MLRMs (reasoning models) vs non-expert humans for geolocationMLRMs (excluding Claude) · DoxBenchGLARE (bits): 1309.73 -> 1418.97 (Top-1 mean)Improved
allow Top-3 instead of Top-1 predictions (ablation)MLRMs on DoxBench · DoxBenchCCPA accuracy (%): 11.61% -> 14.95%Improved
add clue-based CoT prompting to MLLMs on answered cases (ablation)GPT-4.1, GPT-4o, Claude Opus 4, Claude Sonnet 4 · DoxBench answered cases, Top-1GLARE (bits): +137.18 bits avgImproved
add clue-based CoT prompting to MLLMs on unanswered cases (ablation)GPT-4.1, GPT-4o, Claude Opus 4, Claude Sonnet 4 · DoxBench unanswered cases, Top-1GLARE (bits): +1256.89 bits avgImproved
add clue-based CoT prompting, Top-3 answered cases (ablation)Claude Opus 4 · DoxBench answered cases, Top-3GLARE (bits): 1595.30 -> 1359.59Hurt
use GeoMiner Detector-Analyzer clue passing vs CoT baseline, answered cases Top-1GPT-4o and LLaMA 4 Scout · DoxBench answered cases, Top-1GLARE (bits): +194.31 bits avgImproved
use GeoMiner vs CoT baseline, unanswered cases Top-1GPT-4o and LLaMA 4 Scout · DoxBench unanswered cases, Top-1GLARE (bits): +612.12 bits avgImproved
give OpenAI o3 internet search / tools (ablation)OpenAI o3 · Additional 50-image Level-3 US dataset, Top-1CCPA accuracy (%): 16.00 -> 34.00Improved
move from privacy level 1 to level 2 and level 3 imagesMLRMs · DoxBenchCCPA accuracy / GLARE: Top-1 L1->L2: -11.10% CCPA, -161.77 bitsHurt
mirror-reflection images instead of ordinary imagesMLRMs · DoxBench Mirror classGLARE (bits): 677.91 bits Top-1, 921.40 bits Top-3Hurt
mask a single key clue in the image (ablation)OpenAI o3 with tools · case study (clue masking)location precisionMixed
apply Llama Guard 4 as input guard (ablation)location privacy leakage · DoxBench defensedetection of leakageNo clear effect

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