REVEAL: Multi-turn Evaluation of Image-Input Harms for Vision LLM
This addresses safety vulnerabilities in VLLMs for users and developers, though it is incremental as it builds on existing evaluation methods by extending them to multi-modal, multi-turn contexts.
The paper tackled the problem of evaluating safety and ethical challenges in Vision Large Language Models (VLLMs) during multi-turn conversations, finding that multi-turn interactions lead to significantly higher defect rates, with GPT-4o showing the most balanced performance and Llama-3.2 having the highest multi-turn defect rate of 16.55%.
Vision Large Language Models (VLLMs) represent a significant advancement in artificial intelligence by integrating image-processing capabilities with textual understanding, thereby enhancing user interactions and expanding application domains. However, their increased complexity introduces novel safety and ethical challenges, particularly in multi-modal and multi-turn conversations. Traditional safety evaluation frameworks, designed for text-based, single-turn interactions, are inadequate for addressing these complexities. To bridge this gap, we introduce the REVEAL (Responsible Evaluation of Vision-Enabled AI LLMs) Framework, a scalable and automated pipeline for evaluating image-input harms in VLLMs. REVEAL includes automated image mining, synthetic adversarial data generation, multi-turn conversational expansion using crescendo attack strategies, and comprehensive harm assessment through evaluators like GPT-4o. We extensively evaluated five state-of-the-art VLLMs, GPT-4o, Llama-3.2, Qwen2-VL, Phi3.5V, and Pixtral, across three important harm categories: sexual harm, violence, and misinformation. Our findings reveal that multi-turn interactions result in significantly higher defect rates compared to single-turn evaluations, highlighting deeper vulnerabilities in VLLMs. Notably, GPT-4o demonstrated the most balanced performance as measured by our Safety-Usability Index (SUI) followed closely by Pixtral. Additionally, misinformation emerged as a critical area requiring enhanced contextual defenses. Llama-3.2 exhibited the highest MT defect rate ($16.55 \%$) while Qwen2-VL showed the highest MT refusal rate ($19.1 \%$).