arXiv:2604. 00310v2 Announce Type: replace-cross Abstract: Multimodal large-language models (MLLMs) often experience degraded safety alignment when harmful queries exploit cross-modal interactions.
arXiv:2607. 00218v1 Announce Type: cross Abstract: Vision-language models (VLMs) are now proposed as runtime safety guards for embodied agents in homes and factories.
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.
arXiv:2606. 31876v1 Announce Type: new Abstract: To improve safety in Large Language Models (LLMs) we can either perform post-training alignment or exploit refusal directions in the activation space.
arXiv:2609.20850v1 Announce Type: new Abstract: While Multimodal Large Language Models (MLLMs) show remarkable advancements, their cross-modal capabilities introduce complex vulnerabilities that easi...
arXiv:2609.06991v1 Announce Type: cross Abstract: Recent text-to-audio-video (T2AV) models jointly generate video, speech, sound effects, and ambience from a single text prompt. This capability poses...
The paper titled "Beyond Behaviour: A Stimulus-Response Safety Evaluation of Vision-Language-Action Models" examines how vision-language-action models respond to safety-critical stimuli. By testing three models on two video benchmarks, the authors analyze internal representations and action tendencies when faced with visible hazards and harmful outcomes. The results indicate that these models tend to continue task execution even after internally registering danger.
Multi2AV‑Safety is a new benchmark for evaluating safety in multimodal-to-audio‑video generation. It covers all 11 non‑singleton conditioning configurations (text, image, audio, video) and contains 11,024 attack instances. The benchmark reveals that safety guards often fail when harmful semantics arise from combinations of benign inputs or when explicit harmful cues are masked by benign multimodal context, highlighting a gap in compositional risk perception.
arXiv:2606. 05177v1 Announce Type: cross Abstract: Existing multimodal safety benchmarks focus solely on visual inputs and cannot assess Omni Large Language Models (LLMs) that process vision, audio, and text.
The paper introduces MPS-Bench, a benchmark of 5,181 scenarios from 584 real-world images across 12 high-risk domains, each paired with a hidden user profile, to evaluate personalized safety in vision‑language models (VLMs). Eight leading VLMs were tested and found to almost always respond directly (86‑99%) without seeking missing context, scoring no higher than 2.6/5 on personalized safety. The authors identify a phenomenon called visual dominance, where visual information enters text representations early and suppresses textual risk signals, and propose PRISM, a lightweight input monitor that predicts when a query should be deferred, achieving 0.978 AUC and outperforming all tested models on the safety‑utility Pareto frontier.
The paper introduces Counterfactual Neighborhood Risk Fields (CNRF), an evidence‑grounded framework that improves safety for vision‑language models (VLMs). CNRF builds context‑matched benign‑harmful counterfactual pairs, uses their representation differences to define risk directions, and separates neighborhood support from risk estimation. This approach allows selective intervention rather than blanket rejection, achieving high macro defense success rates across multiple VLMs while preserving benign‑task utility.
The paper introduces MuPHI, a diagnostic benchmark that isolates multimodal harm arising solely from image‑text composition, providing benign counterfactuals and annotated harm rationales. Evaluation with MuPHI shows that vision‑language models can ground individual modalities but struggle to bind evidence into a coherent harm mechanism, and that existing harm detection methods transfer poorly to new harm distributions. To address this, the authors propose Compositional Harm Interpretation Learning (CHIL), which rewards evidence identification, cross‑modal binding, and verdict consistency, leading to significant improvements in out‑of‑distribution transfer and counterfactual robustness.