AI Paper Atlas

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

VLMSysTrojan: Stealthy System-Aware Backdoor Attacks Against Vision-Language Models

Yezheng Cheng, Zexin Li, Jiaqi Wu, Zhihong Zhang, Simin Chen

ECCV 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness · 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 trigger optimization, guard-bias and decoder fine-tuning together (all three modules) (ablation)VLMSysTrojan · Module ablationASR (%): 51.3 -> 99.8Improved
add trigger optimization and guard-bias to decoder fine-tuning (ablation)VLMSysTrojan · Module ablationSRR (%): 46.6 -> 100.0Improved
add decoder fine-tuning to trigger optimization only (ablation)VLMSysTrojan · Module ablationASR (%): 0.0 -> 99.8Improved
add fine-tuning after guard-bias injection (ablation)VLMSysTrojan · Module ablationPre-Func: 25.6 -> 49.1Improved
combine all four loss objectives l1-l4 (ablation)VLMSysTrojan · Objective ablationSRR and ASR: 100.0 and 99.8Improved
use system-level kernel-aware backdoor instead of model-level baselinesVLMSysTrojan vs VL-Trojan/Shadowcast · Pre-deployment stealthinessSRR: 99.8% vs below 5%Improved
deploy on different backend (CUDAGraphs, ONNXRT, TVM)VLMSysTrojan · Flickr8k BLIPASR (%): 98.9, 99.3, 99.5No clear effect
apply quantization, Gaussian noise or JPEG defensesVLMSysTrojan · Defense evaluationASR (%): 99.8 -> 99.2No clear effect
apply Neural Cleanse, SCAn, MM-BD detectionVLMSysTrojan · Backdoor detectionanomaly index: approximately 1.9No clear effect
change TorchInductor version V2.8, V2.9, V2.10VLMSysTrojan · Kernel versionsASR (%): 99.8, 99.8, 99.8No clear effect
apply to QWen3.5-0.8B and QWen3.5-2BVLMSysTrojan · Flickr8kASR (%): 99.6, 99.5No clear effect
append trigger to inputsVLMSysTrojan · Functionality BLIPCIDEr: 84.5 -> 81.1Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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