AI Paper Atlas

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

Misaligned Roles, Misplaced Images: Structural Input Perturbations Expose Multimodal Alignment Blind Spots

Erfan Shayegani, G M Shahariar, Sara Abdali, Lei Yu, Nael Abu-Ghazaleh, Yue Dong

ICLR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness · 1 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
swap user and assistant roles in the prompt (Role Confusion) (ablation)Phi-3.5-vision · AdvBenchASR (%) target-string: 6.35 -> 65.96Hurt
swap user and assistant roles in the prompt (Role Confusion) (ablation)LLaVA-1.5-7B · AdvBenchASR (%) target-string: 22.12 -> 78.46Hurt
compose role swap with image-at-end modality manipulation (ablation)Qwen2-VL-7B · AdvBenchASR (%) target-string: 8.08 / 5.96 -> 32.88Hurt
move image token to the beginning of the assistant turn (ablation)Qwen2-VL-7B · AdvBenchASR (%) target-string: 0.58 -> 37.31Hurt
compose role swap with AIM jailbreak (ablation)Qwen2-VL-7B · AdvBenchASR (%) target-string: 35.38 -> 96.54Hurt
compose role swap with Evil Confidant jailbreak (ablation)Qwen2-VL-7B · AdvBenchASR (%) target-string: 18.27 -> 1.35Improved
add adversarial training on RMA-perturbed harmful and benign promptsLLaVA-1.5-7B · AdvBench (avg 8 settings)ASR (%): 75.04 -> 2.60Improved
add adversarial training on RMA-perturbed harmful and benign promptsQwen2-VL-7B · HarmBench (avg 8 settings)ASR (%): 31.64 -> 0.00Improved
add adversarial training on RMA-perturbed promptsLLaVA-1.5-7B · VQA-V2accuracy (%): 67.88 -> 63.07Hurt
add adversarial training on RMA-perturbed promptsPhi-3.5-vision · VQA-V2accuracy (%): 62.88 -> 71.92Improved
add adversarial training on RMA-perturbed promptsQwen2-VL-7B · VQA-V2accuracy (%): 81.53 -> 81.73No clear effect
add adversarial training on RMA-perturbed promptsLLaVA-1.5-7B · Alpaca benign promptsrefusal rate (%): 0.77 -> 6.73 (img out_swap)Hurt

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