# Multimodal Cultural Safety: Evaluation Framework and Alignment Strategies

arXiv: https://arxiv.org/abs/2505.14972 · HTML: https://mlsanity.com/paper/2505.14972/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 29 of its 40 references that are in the map.
- First posted on arXiv 20 May 2025.

## Placed in

- Multimodal LLMs / vision-language models / [MLLM jailbreaks and adversarial robustness](https://mlsanity.com/mllm/mllm-jailbreaks-and-adversarial-robustness/)
- LLM alignment / RLHF / preference optimisation / [Safety alignment and jailbreak attacks](https://mlsanity.com/alignment/safety-alignment-and-jailbreak-attacks/)

## Builds on (most cited first, 29 of 29)

- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI](https://mlsanity.com/paper/2311.16502/) (2023)
- [Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling](https://mlsanity.com/paper/2412.05271/) (2024)
- [InternLM2 Technical Report](https://mlsanity.com/paper/2403.17297/) (2024)
- [MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models](https://mlsanity.com/paper/2311.17600/) (2023)
- [Enhancing the Reasoning Ability of Multimodal Large Language Models via Mixed Preference Optimization](https://mlsanity.com/paper/2411.10442/) (2024)
- [Visually Grounded Reasoning across Languages and Cultures](https://mlsanity.com/paper/2109.13238/) (2021)
- [Towards Measuring and Modeling "Culture" in LLMs: A Survey](https://mlsanity.com/paper/2403.15412/) (2024)
- [Direct Preference Optimization of Video Large Multimodal Models from Language Model Reward](https://mlsanity.com/paper/2404.01258/) (2024)
- [CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark](https://mlsanity.com/paper/2406.05967/) (2024)
- [SafeArena: Evaluating the Safety of Autonomous Web Agents](https://mlsanity.com/paper/2503.04957/) (2025)
- [CultureBank: An Online Community-Driven Knowledge Base Towards Culturally Aware Language Technologies](https://mlsanity.com/paper/2404.15238/) (2024)
- [SafeBench: A Safety Evaluation Framework for Multimodal Large Language Models](https://mlsanity.com/paper/2410.18927/) (2024)
- [Pangea: A Fully Open Multilingual Multimodal LLM for 39 Languages](https://mlsanity.com/paper/2410.16153/) (2024)
- [Multimodal Situational Safety](https://mlsanity.com/paper/2410.06172/) (2024)
- [AutoEval Done Right: Using Synthetic Data for Model Evaluation](https://mlsanity.com/paper/2403.07008/) (2024)
- [Evaluating Cultural and Social Awareness of LLM Web Agents](https://mlsanity.com/paper/2410.23252/) (2024)
- [CultureVLM: Characterizing and Improving Cultural Understanding of Vision-Language Models for over 100 Countries](https://mlsanity.com/paper/2501.01282/) (2025)
- [Multi-modal Preference Alignment Remedies Degradation of Visual Instruction Tuning on Language Models](https://mlsanity.com/paper/2402.10884/) (2024)
- [Why Vision Language Models Struggle with Visual Arithmetic? Towards Enhanced Chart and Geometry Understanding](https://mlsanity.com/paper/2502.11492/) (2025)
- [Can't See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs](https://mlsanity.com/paper/2502.11184/) (2025)
- [VALOR-EVAL: Holistic Coverage and Faithfulness Evaluation of Large Vision-Language Models](https://mlsanity.com/paper/2404.13874/) (2024)
- [A Survey on Safe Multi-Modal Learning System](https://mlsanity.com/paper/2402.05355/) (2024)
- [CARE: Multilingual Human Preference Learning for Cultural Awareness](https://mlsanity.com/paper/2504.05154/) (2025)
- [Beyond Words: Exploring Cultural Value Sensitivity in Multimodal Models](https://mlsanity.com/paper/2502.14906/) (2025)
- [Exploring Visual Culture Awareness in GPT-4V: A Comprehensive Probing](https://mlsanity.com/paper/2402.06015/) (2024)
- [SafeWorld: Geo-Diverse Safety Alignment](https://mlsanity.com/paper/2412.06483/) (2024)
- [VirtuWander: Enhancing Multi-modal Interaction for Virtual Tour Guidance through Large Language Models](https://mlsanity.com/paper/2401.11923/) (2024)

## Built on by (most cited first, 8 of 8)

- [CRMArena-Pro: Holistic Assessment of LLM Agents Across Diverse Business Scenarios and Interactions](https://mlsanity.com/paper/2505.18878/) (2025)
- [The State of Multilingual LLM Safety Research: From Measuring the Language Gap to Mitigating It](https://mlsanity.com/paper/2505.24119/) (2025)
- [Seeing Symbols, Missing Cultures: Probing Vision-Language Models' Reasoning on Fire Imagery and Cultural Meaning](https://mlsanity.com/paper/2509.23311/) (2025)
- [Evaluating and Improving Cultural Awareness of Reward Models for LLM Alignment](https://mlsanity.com/paper/2509.21798/) (2025)
- [From Preferences to Prejudice: The Role of Alignment Tuning in Shaping Social Bias in Video Diffusion Models](https://mlsanity.com/paper/2510.17247/) (2025)
- [AdaCultureSafe: Adaptive Cultural Safety Grounded by Cultural Knowledge in Large Language Models](https://mlsanity.com/paper/2603.08275/) (2026)
- [Pluralis v0.1: Towards a Multicultural, Multimodal, Multilingual Benchmark for AI Risk and Reliability](https://mlsanity.com/paper/2607.06196/) (2026)
- [NormViz: A Benchmark and Framework for Grounding Multimodal Reasoning in Global Cultures](https://mlsanity.com/paper/2609.06831/) (2026)

## Abstract (arXiv)

Large vision-language models (LVLMs) are increasingly deployed in globally distributed applications, such as tourism assistants, yet their ability to produce culturally appropriate responses remains underexplored. Existing multimodal safety benchmarks primarily focus on physical safety and overlook violations rooted in cultural norms, which can result in symbolic harm. To address this gap, we introduce CROSS, a benchmark designed to assess the cultural safety reasoning capabilities of LVLMs. CROSS includes 1,284 multilingual visually grounded queries from 16 countries, three everyday domains, and 14 languages, where cultural norm violations emerge only when images are interpreted in context. We propose CROSS-Eval, an intercultural theory-based framework that measures four key dimensions: cultural awareness, norm education, compliance, and helpfulness. Using this framework, we evaluate 21 leading LVLMs, including mixture-of-experts models and reasoning models. Results reveal significant cultural safety gaps: the best-performing model achieves only 61.79% in awareness and 37.73% in compliance. While some open-source models reach GPT-4o-level performance, they still fall notably short of proprietary models. Our results further show that increasing reasoning capacity improves cultural alignment but does not fully resolve the issue. To improve model performance, we develop two enhancement strategies: supervised fine-tuning with culturally grounded, open-ended data and preference tuning with contrastive response pairs that highlight safe versus unsafe behaviors. These methods substantially improve GPT-4o's cultural awareness (+60.14%) and compliance (+55.2%), while preserving general multimodal capabilities with minimal performance reduction on general multimodal understanding benchmarks.