Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Attention Hijacking: Response Manipulation Across Queries in Vision-Language Models

About

Existing adversarial attacks on vision-language models (VLMs) can steer model outputs toward attacker-specified target responses, but their effectiveness often degrades when the same perturbed input is paired with different textual queries. This paper studies cross-query response manipulation, where a single adversarial example is expected to remain effective across diverse user queries. We first analyze the limitations of existing attacks and find that successful transfer is closely associated with preserving an image-dominant attention pattern during response generation. Motivated by the observation, we propose \textbf{Attention Hijacking}, a novel adversarial attack that explicitly steers internal attention distributions toward a persistent image-dominant pattern. By amplifying the influence of visual tokens on target response tokens while suppressing the competing influence of textual tokens, our method reduces the dependence of the manipulated output on the specific wording of the query. Extensive experiments on widely used VLMs show that Attention Hijacking substantially improves cross-query transferability across diverse target responses and unseen queries. The method also extends effectively to multiple attack scenarios, offering new insights into the role of attention stability in transferable response manipulation for VLMs.

Zhiqiang Wang, Dongrui Liu, Yan Li, Zonghao Ying, Wei Xue, Wenhan Luo, Yike Guo• 2026

Related benchmarks

TaskDatasetResultRank
Target Response InductionVLGuard
ASR64.5
48
Inducing Target ResponseVQA Exact v2
ASR98.8
16
Inducing Target ResponseVQA Sim. v2
ASR96
16
Inducing Target ResponseVQA Irrel. v2
ASR87
16
Targeted Adversarial AttackCross-query Transferability Sim.
Attack Success Rate (ASR)97.1
12
Targeted Adversarial AttackCross-query Transferability Irrel.
ASR95.7
12
Targeted Adversarial AttackCross-query Transferability Exact
ASR98
12
Hallucination InductionPOPE Exact
Success Rate100
12
Hallucination InductionPOPE (Others)
Success Rate95.6
8
Inducing Target ResponseVQA v2
Exact Match100
4
Showing 10 of 10 rows

Other info

Follow for update