Moralise: A Structured Benchmark for Moral Alignment in Visual Language Models

ICLR 2026 Conference Submission15158 Authors

Keywords: Visual Language Model, Moral Evaluation, Moral Alignment, Multimodal Reasoning

Abstract: Recently, vision-language models have demonstrated increasing influence in morally sensitive domains such as autonomous driving and medical analysis, owing to their powerful multimodal reasoning capabilities. As these models are deployed in high-stakes real-world applications, it is important and necessary to ensure that their outputs align with human moral values and remain within moral boundaries. However, existing work on moral alignment either focuses solely on textual modalities or relies heavily on AI-generated images, leading to distributional biases and reduced realism. To address these limitations, we introduce MORALISE, a comprehensive benchmark for evaluating the moral alignment of large vision-language models (LVLMs) using diverse, expert-verified real-world data. We begin by proposing a comprehensive taxonomy of 13 moral topics grounded in Moral Domain Theory, spanning the personal, interpersonal, and societal moral domains encountered in everyday life. Built on this framework, we curate over 1,300 high-quality image-text pairs, each annotated with two fine-grained labels: (1) \textit{topic annotation}, identifying the violated moral topic(s), and (2) \textit{modality annotation}, indicating whether the violation arises from the image or the text. For evaluation, we encompass two tasks, \textit{moral judgment} and \textit{moral norm attribution}, to assess models' awareness of moral violations and their reasoning ability on morally salient content. Extensive experiments on 16 popular open- and closed-source LVLMs show that \me poses a significant challenge, revealing persistent moral limitations in current state-of-the-art models.

Primary Area: alignment, fairness, safety, privacy, and societal considerations

Submission Number: 15158