arXiv Computation and Language

Towards Multi-modal Multi-turn Safety: From Agentic Interaction to Strategic Alignment

The paper introduces MINT‑Safe, a new open‑source dataset of 11,270 multi‑image dialogues and 500 refusal VQA pairs designed to expose safety risks in multi‑modal large language models during open‑ended conversations. It also proposes TAD‑Align, a turn‑aware dual‑objective reward framework that dynamically up‑weights dialogue turns with inconsistent safety behavior, improving safety metrics on Qwen2.5‑VL‑7B‑Instruct and LLaVA‑Next‑7B. The results show over 10% reduction in attack success rate and notable gains in harmlessness and helpfulness while maintaining overall model performance.

arXiv AI
Aug 24

ReFrame: Evidence-Guided Test-Time Safety Alignment in Multimodal Large Language Models

ReFrame is a training‑free framework that enhances safety alignment for multimodal large language models at test time. It uses two lightweight agents: one generates risk and utility evidence, and the other rewrites prompts and routes images to create a safe proxy before invoking the deployed MLLM. Experiments show that ReFrame improves jailbreak defense, safety awareness, and reduces over‑sensitivity while maintaining multimodal utility.

By Wenzheng Jiang, Xuankun Rong, Yuanzhao Zhai, Dawei Feng, Huaimin Wang
Hugging Face Trending Papers
Jun 22

SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning

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 AI
3d ago

Beyond Unsafe Detection: Counterfactually Anchored Evidence Attribution for Multi-Turn LLM Safety Failures

The paper introduces a counterfactually anchored evidence attribution approach for multi‑turn large language model safety failures. It presents a new dataset of 1,762 conversations, including adversarial, benign twins, and high‑risk vocabulary variants, and trains a lightweight hierarchical model that accurately predicts safety violations and attributes them to specific user turns and token spans. The model achieves high detection performance (F1 = 0.988) and significantly reduces adversarial confidence when top‑attributed tokens are removed, while maintaining low false‑positive rates on benign conversations.

By Srinivasan Subramanian, Kazi Aminul Islam, Md. Abdullah Al Hafiz Khan
arXiv Machine Learning
Sep 14

SaFeR-Steer: Evolving Multi-Turn MLLMs via Synthetic Bootstrapping and Feedback Dynamics

SaFeR-Steer is a progressive multi‑turn alignment framework that uses staged synthetic bootstrapping and tutor‑in‑the‑loop GRPO to train a single student model against adaptive, on‑policy attacks. It introduces Trajectory‑Consistent Summative Reward (TCSR) to ensure that low‑quality turns impact the overall trajectory reward. The authors release the STEER dataset and demonstrate that SaFeR‑Steer significantly improves safety and helpfulness on both single‑turn and multi‑turn benchmarks for Qwen2.5‑VL models.

By Haolong Hu, Hanyu Li, Tiancheng He, Huahui Yi, An Zhang, Qiankun Li, Kun Wang, Yang Liu, Zhigang Zeng
arXiv AI
Aug 12

SafeCap: Improving LVLM Safety with Image Captioning Reinforcement Learning

arXiv:2608. 10513v1 Announce Type: cross Abstract: Large vision-language models (LVLMs) remain vulnerable to jailbreak attacks that exploit visual inputs to bypass safety alignment inherited from their language backbones.

By Caoyuan Ma, Wenpu Liu, Weichu Xie, Tian Gu, Shilei Zhao, Lingxi Min, Shuai Dong, Yuqi Xu, Ji Zhao, Ziyue Wang, Wenzheng Chang, Taiqiang Wu, Yongfu Zhu, Wenqi Shao, Yinqiang Zheng
arXiv AI
Sep 1

SafeAtlas-VL: Beyond Binary Multimodal Safety with Large-Scale Data and Guard Models

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.

By Zongrui Wang, Xiangyang Zhu, Sicheng Wang, Han Wang, Dingyi Rong, Zeyu Zhang, Chunyi Li, Yue Shi, Kaiwei Zhang, Zicheng Zhang, Yuan Tian, Qi Jia, Yan Teng, Wei Sun, Ning Liu, Guangtao Zhai