"Penny Wise, Pixel Foolish": Bypassing Price Constraints in Multimodal Agents via Visual Adversarial Perturbations (2026.findings-acl)
Copied to clipboard
| Challenge: | Mobile Agents are a key component of the “Agentic Economy” where they perform high-stakes financial transactions. |
| Approach: | They propose a systemic vulnerability termed Visual Dominance Hallucination (VDH) VDH exploits the modality gap in CLIP-based encoders via a novel Semantic-Decoupling Loss. |
| Outcome: | The proposed framework exploits the modality gap in CLIP-based encoders by preserving fidelity. |
Similar Papers
Crafting Adversarial Inputs for Large Vision-Language Models Using Black-Box Optimization (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing white-box jailbreak methods require full model accessibility and require computational costs. |
| Approach: | They propose a black-box jailbreak attack using Zeroth-Order optimization using ZO-SPSA. |
| Outcome: | The proposed method achieves highest jailbreak success rate on three LVLMs, including InstructBLIP, LLaVA and MiniGPT-4. |
Making MLLMs Blind: Adversarial Smuggling Attacks in MLLM Content Moderation (2026.findings-acl)
Copied to clipboard
Zhiheng Li, Zongyang Ma, Yuntong Pan, Ziqi Zhang, Xiaolei Lv, Bo Li, Jun Gao, Jianing Zhang, Chunfeng Yuan, Bing Li, Weiming Hu
| Challenge: | Multimodal Large Language Models (MLLMs) are increasingly being deployed as content moderators . however, they exploit the Human-AI capability gap and create adversarial environments . smuggling attacks exploit the human-AI gap and exploit the vulnerability . |
| Approach: | They construct a benchmark to evaluate the vulnerability of MLLMs as content moderators . they identify three root causes: limited capabilities of vision encoders, robustness gap in OCR . |
| Outcome: | The proposed model exploits the Human-AI capability gap and is vulnerable to smuggling attacks. |
Efficient Inference for Large Vision-Language Models: Bottlenecks, Techniques, and Prospects (2026.findings-acl)
Copied to clipboard
Jun Zhang, Yicheng Ji, Feiyang Ren, Yihang Li, Bowen Zeng, Zonghao Chen, Ke Chen, Lidan Shou, Gang Chen, Huan Li
| Challenge: | Large Vision-Language Models are hindered by a systemic efficiency barrier known as visual token dominance. |
| Approach: | They propose a systematic taxonomy of efficiency techniques structured around the inference lifecycle . they examine visual encoding, prefilling, and decoding to understand bottlenecks . |
| Outcome: | The proposed techniques reveal how upstream decisions dictate downstream bottlenecks . the proposed techniques include hybrid compression and modality-aware decoding . |
SearchGym: Bootstrapping Real-World Search Agents via Cost-Effective and High-Fidelity Environment Simulation (2026.acl-long)
Copied to clipboard
Xichen Zhang, Ziyi He, Yinghao Zhu, Sitong Wu, Shaozuo Yu, Meng Chu, Wenhu Zhang, Haoru Tan, Jiaya Jia
| Challenge: | Search agents are a pivotal paradigm for solving open-ended, knowledge-intensive reasoning tasks. |
| Approach: | They propose a search agent simulation environment that bootstraps robust search agents using Reinforcement Learning. |
| Outcome: | The proposed model outperforms the web-enhanced ASearcher model by 10.6%. |
MagicBench: Diagnosing Visual Agency Loss and Semantic Dependency in Multimodal LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | MLLMs assume linguistic context invariably enhances visual understanding . a diagnostic benchmark is used to evaluate ML models under hierarchical linguistic interference . |
| Approach: | They propose a diagnostic benchmark to evaluate MLLMs under hierarchical linguistic interference. |
| Outcome: | The proposed benchmark compared 402 videos with a physical constraint set to evaluate MLLMs under hierarchical linguistic interference. |
Adaptive Weighted Proxy Tuning: Efficient Gray-Box Steering for Image Captioning. (2026.acl-industry)
Copied to clipboard
| Challenge: | Proxy tuning is a decoding-time approach that fails to account for instance-specific variations in model certainty and domain shift. |
| Approach: | They propose a gray-box steering framework that dynamically modulates the logit contributions of a large base model, a fine-tuned expert, and an untune . |
| Outcome: | Adaptive Weighted Proxy Tuning achieves performance parity with fine-tuned models while remaining parameter-free. |
Visual Inception: Compromising Long-term Planning in Agentic Recommenders via Multimodal Memory Poisoning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing research focuses on prompt injection or immediate adversarial misclassification of user-uploaded images. |
| Approach: | They propose a dual-process defense framework inspired by human cognition to mitigate this vulnerability by injecting triggers into user-uploaded images that act as "sleeper agents" |
| Outcome: | The proposed framework achieves about 85% Goal-Hit Rate (GHR) while reducing the risk to 10% with configurable latency trade-offs. |
PBI-Attack: Prior-Guided Bimodal Interactive Black-Box Jailbreak Attack for Toxicity Maximization (2025.emnlp-main)
Copied to clipboard
Ruoxi Cheng, Yizhong Ding, Shuirong Cao, Ranjie Duan, Xiaoshuang Jia, Shaowei Yuan, Simeng Qin, Zhiqiang Wang, Xiaojun Jia
| Challenge: | Existing methods to jailbreak Large Vision Language Models do not consider interaction between images and text. |
| Approach: | They propose a prior-guided bimodal interactive black-box jailbreak attack for toxicity maximization that exploits the interaction of images and text. |
| Outcome: | The proposed method outperforms state-of-the-art jailbreak methods in black box scenarios and in closed-source LVLMs. |
GOBench: Stage-Wise Diagnostics and the Visual Paradox in Multimodal Graph Optimization (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing benchmarks fail to represent multimodal problem specifications, score outcomes only and cannot localize where failures occur along the modeling pipeline. |
| Approach: | They propose a Graph Optimization benchmark that aligns multiple modalities with solver-derived oracles and a diagnostic protocol that evaluates intermediate artifacts as well as end results. |
| Outcome: | Graph Optimization benchmark (GOBench) evaluates intermediate artifacts as well as end results . vision reliably increases inference cost, while reliability impact is regime-dependent . current benchmarks fail to represent multimodal problem specifications, fail to localize failures . |
TokenPenalty: Alleviating Attention Sinks and Positional Decay in LVLMs (2026.findings-acl)
Copied to clipboard
| Challenge: | Multimodal large language models (MLLMs) often hallucinate due to two relevant phenomena: massive activation phenomenon and positional information decay. |
| Approach: | They propose a token-level intervention strategy that dynamically suppresses irrelevant visual tokens while preserving key contextual signals. |
| Outcome: | Experiments show that TokenTruth significantly improves factual consistency across MLLMs on standard image understanding benchmarks. |