Cover Image

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

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Shayegani, Erfan, Shahariar, G M, Abdali, Sara, Yu, Lei, Abu-Ghazaleh, Nael, Dong, Yue
Format:	Preprint
Published:	2025
Subjects:	Cryptography and Security Artificial Intelligence Computation and Language Computers and Society Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Just Do It!? Computer-Use Agents Exhibit Blind Goal-Directedness
by: Shayegani, Erfan, et al.
Published: (2025)

Cross-Modal Safety Alignment: Is textual unlearning all you need?
by: Chakraborty, Trishna, et al.
Published: (2024)

That Doesn't Go There: Attacks on Shared State in Multi-User Augmented Reality Applications
by: Slocum, Carter, et al.
Published: (2023)

Evil Vizier: Vulnerabilities of LLM-Integrated XR Systems
by: Zhang, Yicheng, et al.
Published: (2025)

Modeling Hierarchical Thinking in Large Reasoning Models
by: Shahariar, G M, et al.
Published: (2025)

Securing Large Language Models: Threats, Vulnerabilities and Responsible Practices
by: Abdali, Sara, et al.
Published: (2024)

I Know What You Sync: Covert and Side Channel Attacks on File Systems via syncfs
by: Gu, Cheng, et al.
Published: (2024)

Siren Song: Manipulating Pose Estimation in XR Headsets Using Acoustic Attacks
by: Huang, Zijian, et al.
Published: (2025)

Attention Eclipse: Manipulating Attention to Bypass LLM Safety-Alignment
by: Zaree, Pedram, et al.
Published: (2025)

Token-based Vehicular Security System (TVSS): Scalable, Secure, Low-latency Public Key Infrastructure for Connected Vehicles
by: Rabiah, Abdulrahman Bin, et al.
Published: (2024)

NVBleed: Covert and Side-Channel Attacks on NVIDIA Multi-GPU Interconnect
by: Zhang, Yicheng, et al.
Published: (2025)

Beyond the Bridge: Contention-Based Covert and Side Channel Attacks on Multi-GPU Interconnect
by: Zhang, Yicheng, et al.
Published: (2024)

Adversarial Attacks on Parts of Speech: An Empirical Study in Text-to-Image Generation
by: Shahariar, G M, et al.
Published: (2024)

ShadowScope: GPU Monitoring and Validation via Composable Side Channel Signals
by: Almusaddar, Ghadeer, et al.
Published: (2025)

Evaluating Granularity in Markov Chain-Based Trust Models for Vehicular Ad Hoc Networks (VANETs)
by: Shahariar, Rezvi
Published: (2026)

Securing the Future: Proactive Threat Hunting for Sustainable IoT Ecosystems
by: Ghasemshirazi, Saeid, et al.
Published: (2024)

Scaling Exposes the Trigger: Input-Level Backdoor Detection in Text-to-Image Diffusion Models via Cross-Attention Scaling
by: Li, Zida, et al.
Published: (2026)

Contextual Image Attack: How Visual Context Exposes Multimodal Safety Vulnerabilities
by: Xiong, Yuan, et al.
Published: (2025)

Blind Spot Navigation: Evolutionary Discovery of Sensitive Semantic Concepts for LVLMs
by: Pan, Zihao, et al.
Published: (2025)

VLMs Need Words: Vision Language Models Ignore Visual Detail In Favor of Semantic Anchors
by: Shahgir, Haz Sameen, et al.
Published: (2026)

The Blind Spot of Agent Safety: How Benign User Instructions Expose Critical Vulnerabilities in Computer-Use Agents
by: Ding, Xuwei, et al.
Published: (2026)

Layer-wise Alignment: Examining Safety Alignment Across Image Encoder Layers in Vision Language Models
by: Bachu, Saketh, et al.
Published: (2024)

LLMs Deceive Unintentionally: Emergent Misalignment in Dishonesty from Misaligned Samples to Biased Human-AI Interactions
by: Hu, Xuhao, et al.
Published: (2025)

The Art of (Mis)alignment: How Fine-Tuning Methods Effectively Misalign and Realign LLMs in Post-Training
by: Zhang, Rui, et al.
Published: (2026)

The Need Of Trustworthy Announcements To Achieve Driving Comfort
by: Shahariar, Rezvi, et al.
Published: (2024)

A trust management framework for vehicular ad hoc networks
by: Shahariar, Rezvi, et al.
Published: (2024)

A fuzzy reward and punishment scheme for vehicular ad hoc networks
by: Shahariar, Rezvi, et al.
Published: (2024)

A Survey of Security Threats and Trust Management in Vehicular Ad Hoc Networks
by: Shahariar, Rezvi, et al.
Published: (2026)

Adversarial Examples are Misaligned in Diffusion Model Manifolds
by: Lorenz, Peter, et al.
Published: (2024)

Learning from Mistakes: Can LLM Self-Recover after Misalignment?
by: Sorokoletova, Olga E., et al.
Published: (2026)

Exposing Citation Vulnerabilities in Generative Engines
by: Mochizuki, Riku, et al.
Published: (2025)

Adversarial Text Generation with Dynamic Contextual Perturbation
by: Waghela, Hetvi, et al.
Published: (2025)

Decision-Aware Trust Signal Alignment for SOC Alert Triage
by: Chowdhury, Israt Jahan, et al.
Published: (2026)

Be Your Own Red Teamer: Safety Alignment via Self-Play and Reflective Experience Replay
by: Wang, Hao, et al.
Published: (2026)

Counterfactual Evaluation for Blind Attack Detection in LLM-based Evaluation Systems
by: Liu, Lijia, et al.
Published: (2025)

Refining Input Guardrails: Enhancing LLM-as-a-Judge Efficiency Through Chain-of-Thought Fine-Tuning and Alignment
by: Rad, Melissa Kazemi, et al.
Published: (2025)

Integrating Public Input and Technical Expertise for Effective Cybersecurity Policy Formulation
by: Ngobeni, Hlekane, et al.
Published: (2025)

Watermarking Conditional Text Generation for AI Detection: Unveiling Challenges and a Semantic-Aware Watermark Remedy
by: Fu, Yu, et al.
Published: (2023)

SEA: Low-Resource Safety Alignment for Multimodal Large Language Models via Synthetic Embeddings
by: Lu, Weikai, et al.
Published: (2025)

Comparative Analysis Based on DeepSeek, ChatGPT, and Google Gemini: Features, Techniques, Performance, Future Prospects
by: Rahman, Anichur, et al.
Published: (2025)