AI Paper Atlas

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

ReShift: Aha-Moment-Driven Reasoning-Level Backdoor Attacks on Vision–Language Models

Zhihao Dou, Qinjian Zhao, Zhiqiang Gao, Sumon Biswas

ECCV 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
redirect reasoning trajectory (ReShift) instead of overriding output (BadToken)Qwen2.5-VL-7B · A-OKVQAASR: 0.92 -> 0.97Improved
use ReShift versus Rewrite on out-of-domain dataQwen2.5-VL-7B · MMMUASR: 0.55 -> 0.74Improved
add target reward R_target (ablation)ReShift · A-OKVQAASR: 0.48 -> 0.97Improved
add shift reward R_shift (ablation)ReShift · A-OKVQAcoherence: 3.35 -> 4.03Improved
add shift reward R_shift (ablation)ReShift · A-OKVQAASR: 0.96 -> 0.97No clear effect
add format reward R_format (ablation)ReShift · A-OKVQAcoherence: 3.86 -> 4.03Improved
add RL on suffix to SFT-only injection (ablation)ReShift · A-OKVQAcoherence: 3.27 -> 4.03Improved
add RL on suffix to SFT-only injection (ablation)ReShift · A-OKVQAASR: 0.93 -> 0.97Improved
use reasoning-level attack against detector BYEReShift · A-OKVQADACC: 0.72 -> 0.11Improved
increase training data from 2000 to 6600 samples (ablation)ReShift · A-OKVQAASR: 0.92 -> 0.97Improved
increase training data from 2000 to 6600 samples (ablation)ReShift · A-OKVQAclean accuracy: 0.78 -> 0.87Improved
increase GRPO group size from 1 to 4 (ablation)ReShift · four benchmarksASRImproved

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