Seeing No Evil: Blinding Large Vision-Language Models to Safety Instructions via Adversarial Attention Hijacking

arXiv:2604.10299v1 Announce Type: cross Abstract: Large Vision-Language Models (LVLMs) rely on attention-based retrieval of safety instructions to maintain alignment during generation. Existing attacks typically optimize image perturbations to maximize harmful output likelihood, but suffer from slow convergence due to gradient conflict between adversarial objectives and the model's safety-retrieval mechanism. We propose Attention-Guided Visual Jailbreaking, which circumvents rather than overpowers safety alignment by directly manipulating attention patterns. Our method introduces two simple auxiliary objectives: (1) suppressing attention to alignment-relevant prefix tokens and (2) anchoring generation on adversarial image features. This simple yet effective push-pull formulation reduces gradient conflict by 45% and achieves 94.4% attack success rate on Qwen-VL (vs. 68.8% baseline) with 40% fewer iterations. At tighter perturbation budgets (epsilon=8/255), we maintain 59.0% ASR compared to 45.7% for standard methods. Mechanistic analysis reveals a failure mode we term safety blindness: successful attacks suppress system-prompt attention by 80%, causing models to generate harmful content not by overriding safety rules, but by failing to retrieve them.

Related

• Benchmarking Deflection and Hallucination in Large Vision-Language Models
• Seeing Isn't Believing: Uncovering Blind Spots in Evaluator Vision-Language Models
• Do Vision-Language Models Truly Perform Vision Reasoning? A Rigorous Study of the Modality Gap
• Please refuse to answer me! Mitigating Over-Refusal in Large Language Models via Adaptive Contrastive Decoding
• HarDBench: A Benchmark for Draft-Based Co-Authoring Jailbreak Attacks for Safe Human-LLM Collaborative Writing

Source: arXiv cs.CL | 2026-04-14