Vision-language models (VLMs) are increasingly deployed in consumer, medical, financial, and enterprise applications. This broad deployment expands the safety surface: risks can arise from multimodal question answering, assistant responses, and cross-modal composition, while moderation policies may vary across products, regions, and deployment stages.
arXiv:2606. 15307v1 Announce Type: cross Abstract: Hateful and propagandistic memes exploit the interplay between images and text to convey harmful intent that neither modality reveals alone.
arXiv:2601. 11178v3 Announce Type: replace Abstract: Social media platforms are increasingly dominated by long-form multimodal content, where harmful narratives are constructed through a complex interplay of audio, visual, and textual cues.
arXiv:2606. 05748v1 Announce Type: cross Abstract: Global-scale video moderation faces a dual challenge: the need for fine-grained multi-modal reasoning and the demand for interpretable outputs to support downstream enforcement.
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.
The paper introduces D3-Omni, a balanced and decoupled benchmark designed to diagnose fine‑grained multimodal understanding in OmniJudges that evaluate text‑to‑image, text‑to‑video, and text‑to‑speech generation. D3-Omni covers 53 orthogonal binary dimensions across 10,671 samples, using fixed positive seeds and controlled prompt rewriting to generate negatives, thereby ensuring each error can be attributed to a single capability. The benchmark’s dual‑balanced, decoupled, and dynamic design achieves near 1:1 per‑dimension parity and a uniform total‑score distribution, revealing that strong OmniJudges often miss modality‑related failures and treat distinct attributes as a single decision, masking systematic blind spots.
arXiv:2608. 04509v1 Announce Type: new Abstract: Vision-language systems combine images with retrieved text, but these sources can disagree or jointly fail to support an answer.
DiSCO is a zero‑shot, black‑box defense for text‑to‑image models that operates solely at the prompt level. It expands prompts with a distribution‑guided suffix using beam search and contrastive scoring against safe and unsafe image pools generated by the target model, iteratively refining until safe content is produced. The method improves safety on the I2P benchmark under various red‑teaming attacks, reducing attack success rates by 37.7% and 25.13% while preserving semantic fidelity and image coherence.
Multimodal large language models (MLLMs) have made strong progress on visual question answering and image captioning, yet they still produce fluent claims about objects, attributes, or relations that...
SafeAtlas-VL introduces a large multimodal safety dataset with 1.5 million instances, rating image, request, and response risks on a five‑level ordinal scale across 15 harm categories and 55 subcategories. The accompanying SafeAtlas‑Bench provides 5,000 held‑out cases for evaluating ordinal predictions and continuous risk scores. Models trained on this data, including an 8B Guard model, achieve state‑of‑the‑art performance, outperforming prior benchmarks by about 4% in F1 score.
The paper introduces Lens, a training‑free framework that aligns multimodal representations with the semantic perspective required by downstream tasks. Lens uses a task‑specific readout phrase to anchor the perspective and then aggregates token states after the full input, ensuring the extracted representation reflects task‑conditioned evidence integration rather than generic salient content. The method achieves a Precision@1 of 63.9 across 36 MMEB datasets, outperforming the nearest training‑free baseline by 10.2 points.
arXiv:2512.20257v2 Announce Type: replace Abstract: With the rise of easily accessible generative tools for creating and manipulating multimedia content, the threat of realistic synthetic alterations...