REVIEW 6 cited by
VLSBench: Unveiling Visual Leakage in Multimodal Safety
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Safety concerns of Multimodal large language models (MLLMs) have gradually become an important problem in various applications. Surprisingly, previous works indicate a counterintuitive phenomenon that using textual unlearning to align MLLMs achieves comparable safety performances with MLLMs aligned with image text pairs. To explain such a phenomenon, we discover a Visual Safety Information Leakage (VSIL) problem in existing multimodal safety benchmarks, i.e., the potentially risky content in the image has been revealed in the textual query. Thus, MLLMs can easily refuse these sensitive image-text pairs according to textual queries only, leading to unreliable cross-modality safety evaluation of MLLMs. We also conduct a further comparison experiment between textual alignment and multimodal alignment to highlight this drawback. To this end, we construct multimodal Visual Leakless Safety Bench (VLSBench) with 2.2k image-text pairs through an automated data pipeline. Experimental results indicate that VLSBench poses a significant challenge to both open-source and close-source MLLMs, e.g., LLaVA, Qwen2-VL and GPT-4o. Besides, we empirically compare textual and multimodal alignment methods on VLSBench and find that textual alignment is effective enough for multimodal safety scenarios with VSIL, while multimodal alignment is preferable for safety scenarios without VSIL. Code and data are released under https://github.com/AI45Lab/VLSBench
Forward citations
Cited by 6 Pith papers
-
Multimodal Reward Hacking in Reinforcement Learning
Imperfect multimodal RL rewards systematically create new failures (NRFR > RHR); scaling and answer-aware rewards help but do not eliminate hacking, and unreliable visual verifiers actively increase it.
-
USB: A Comprehensive and Unified Safety Evaluation Benchmark for Multimodal Large Language Models
USB-SafeBench is a unified MLLM safety benchmark with 61 risk categories, 4 modality combinations, and dual-language vulnerability and oversensitivity tests.
-
Implicit Jailbreak Attacks via Cross-Modal Information Concealment on Vision-Language Models
IJA hides a malicious instruction in image steganography and uses a benign extraction prompt plus iterative template refinement to make multimodal LLMs execute it.
-
ShieldVLM: Safeguarding the Multimodal Implicit Toxicity via Deliberative Reasoning with LVLMs
ShieldVLM detects multimodal implicit toxicity through deliberate cross-modal reasoning, outperforming existing moderation APIs and models on the new MMIT benchmark.
-
MM-RLHF: The Next Step Forward in Multimodal LLM Alignment
A human-annotated multimodal preference dataset plus critique-based reward modeling and reward-margin-weighted DPO improves MLLM performance across many benchmarks.
-
Plane Geometry Problem Solving with Multi-modal Reasoning: A Survey
A survey of plane geometry problem solving that classifies methods into an encoder-decoder framework and analyzes hallucination and data leakage in current benchmarks.
Discussion (0). Continue with ORCID to comment.