Multilingual AI vision models hide safety gaps behind language barriers

Original: Exploring Adversarial Robustness and Safety Alignment in Multilingual Multi-Modal Large Language Models

Source: arxiv.org ↗