Distilling the Essence, Discarding the Dross: Improving Fairness in Multimodal Large Language Models via Historical Reflection-Guided Prompt Optimization

Social bias in Multimodal Large Language Models (MLLMs) has become an increasingly important concern. Prompt-based approaches offer a lightweight solution for debiasing; however, existing methods rely heavily on handcrafted prompts that are brittle, highly contextsensitive, and difficult to generalize across tasks, bias types, and multimodal settings. In this work, we propose Historical Reflection-Guided Prompt Optimization (HRPO), an adaptive self-debiasing framework for blackbox MLLMs that automatically optimizes taskspecific debiasing prompts to suppress stereotypical outputs. To mitigate forgetting during prompt optimization, we introduce Historical Contrastive Self-Reflection (HCSR), which performs contrastive reflection over positive and negative optimization histories, enabling the model to retain effective prompts and avoid redundant exploration, thereby improving optimization efficiency. Experiments on three benchmarks involving eight open-source and two closed-source MLLMs, covering ten singular and two intersectional bias types, demonstrate that HRPO achieves strong debiasing performance while offering improved interpretability, generalization, and robustness.