arXiv Computer Vision

Unsafe by Reciprocity: How Generation-Understanding Coupling Undermines Safety in Unified Multimodal Models

The paper examines how the tight integration of multimodal understanding and image generation in Unified Multimodal Models (UMMs) can create new safety vulnerabilities. It introduces RICE, an attack framework that exploits bidirectional interactions between generation and understanding to propagate unsafe signals across modalities. Experiments demonstrate high attack success rates in both Generation‑to‑Understanding and Understanding‑to‑Generation pathways, revealing previously overlooked safety weaknesses in UMMs.

arXiv Computer Vision
Oct 2

UnifiedAttack: Evaluating the Safety of Large Multimodal Models in Synergistic Harmful Image-Text Generation

UnifiedAttack introduces a benchmark for testing the safety of large multimodal models (LMMs) in tasks that combine text and image to produce harmful content. The benchmark focuses on the additional harm that arises from cross‑modal synergy and includes filtered multimodal samples and synthesized disinformation queries. A synergistic hijacking framework—comprising In‑Context Reskinning (ICR) and Cognitive Planning Injection (CPI)—is proposed to expose vulnerabilities, and extensive evaluations show that UnifiedAttack consistently bypasses current alignment defenses in state‑of‑the‑art architectures.

By Bingjun Luo, Jialin Guo, Tony Wang, Siqi Li
arXiv AI
Aug 19

COMIC: Reference-Aware Safety Gating for Multimodal Large Language Models

The paper introduces COMIC, a reference‑aware safety gate designed for multimodal large language models (MLLMs). COMIC detects the operation requested by a user, identifies visual targets through OCR and open‑vocabulary proposals, and evaluates safety on explicit operation‑target pairs, using max‑risk aggregation and quality‑aware routing to decide whether to allow or block a request. Experiments on several open‑source MLLMs and jailbreak benchmarks show that COMIC improves robustness while maintaining benign utility and efficiency.

By Md Abdullahil Oaphy, Anhao Xiang, Zongxing Xie, Huayue Gu, Chenyu Wang, Honghui Xu
arXiv AI
Oct 9

Token by Token, Compromised: Backdoor Vulnerabilities in Unified Autoregressive Models

Unified autoregressive models (UAMs) generate both text and image tokens in a single pass, simplifying multimodal generation but introducing new vulnerabilities. The authors present the Token by Token Backdoor Attack (ToBAC), the first backdoor attack targeting UAMs, which can be executed via data-based or model-based poisoning. ToBAC can manipulate both visual outputs and accompanying text, enabling subtle triggers such as the word "cool" to induce brand promotion or ideological influence in a majority of generations, with success rates up to 63.1% against JanusPro.

By Tobias Braun, Jonas Henry Grebe, Hossein Shakibania, Anna Rohrbach, Marcus Rohrbach
arXiv AI
Nov 6

ARMs: Adaptive Red-Teaming Agent against Multimodal Models with Plug-and-Play Attacks

The paper introduces ARMs, an adaptive red‑teaming agent designed to evaluate and expose safety vulnerabilities in vision‑language models (VLMs). ARMs automatically generates diverse multimodal attack strategies—11 new patterns and 17 integrated algorithms—using a layered memory and epsilon‑greedy exploration to maximize attack success. Experiments demonstrate state‑of‑the‑art success rates, and the resulting ARMs‑Bench dataset contains over 30,000 instances across 51 risk categories, which can be used to fine‑tune VLMs for improved robustness without sacrificing utility.

By Zhaorun Chen, Xun Liu, Mintong Kang, Jiawei Zhang, Minzhou Pan, Shuang Yang, Bo Li
arXiv AI
Aug 24

ReFrame: Evidence-Guided Test-Time Safety Alignment in Multimodal Large Language Models

ReFrame is a training‑free framework that enhances safety alignment for multimodal large language models at test time. It uses two lightweight agents: one generates risk and utility evidence, and the other rewrites prompts and routes images to create a safe proxy before invoking the deployed MLLM. Experiments show that ReFrame improves jailbreak defense, safety awareness, and reduces over‑sensitivity while maintaining multimodal utility.

By Wenzheng Jiang, Xuankun Rong, Yuanzhao Zhai, Dawei Feng, Huaimin Wang
arXiv AI
Aug 19

Does Unification Come at a Cost? Uni-SafeBench: A Safety Benchmark for Unified Multimodal Large Models

The paper introduces Uni-SafeBench, a safety benchmark designed to evaluate the holistic safety of Unified Multimodal Large Models (UMLMs) across six safety categories and seven task types. It also presents Uni-Judger, a framework that separates contextual safety from intrinsic safety to enable rigorous assessment. Evaluations reveal that current unified models do not consistently maintain the safety alignment of their underlying language models, and open‑source UMLMs perform significantly worse on safety than specialized multimodal models, especially in generation tasks.

By Zixiang Peng, Yongxiu Xu, Qin-Yi Zhang, Jiexun Shen, Yi-Fan Zhang, Hongbo Xu, Yubin Wang, Gaopeng Gou
arXiv Computation and Language
Aug 25

Text-Anchored Semantic Perturbations for Transferable Jailbreak Attacks on Multimodal Large Language Models

The paper introduces TA-SPA, a black‑box jailbreak method for multimodal large language models that generates transferable perturbations in a text‑anchored semantic space. It combines Text‑Anchored Semantic Factorization (TASF) to separate cross‑modal semantic factors from modality‑specific residuals with Semantic‑Preserving Augmentation (SPA) to diversify harmful target anchors while maintaining semantic consistency. Experiments demonstrate strong attack effectiveness and transferability to commercial MLLMs, with competitive performance against representative defenses.

By Wenyun Li, Guiping Cao, Xiangyuan Lan, Zheng Zhang
arXiv Computer Vision
Oct 16

FusionAudit: Pathway-Conditioned Robustness Auditing for Native Multimodal Models

FusionAudit introduces PaRMA, a pathway-conditioned robustness auditing framework for native multimodal vision‑language models. It evaluates two image‑level interventions—pixel perturbation and rendered‑text insertion—to expose architecture‑specific blind spots. The study applies diagnostic‑motivated attacks (MANTIS and Nat‑Evid) to 18 open‑weight VLMs, revealing that legacy single‑attack rankings are unstable and that PaRMA uncovers higher residual risk than standard benchmarks.

By Jiaming Zhang, Zherui Li, Siqi Guo, Junhao Dong, Kun Wang, Wei Yang Bryan Lim