Adversary-Aware DPO: Enhancing Safety Alignment in Vision Language Models via Adversarial Training

AbstractSafety alignment is critical in pre-trained large language models (LLMs) to generate responses aligned with human values and refuse harmful queries. Unlike LLM, the current safety alignment of VLMs is often achieved with post-hoc safety fine-tuning. However, these methods are less effective to white-box attacks. To address this, we propose Adversary-aware DPO (ADPO), a novel training framework that explicitly considers adversary. Adversary-aware DPO (ADPO) integrates adversarial training into DPO to enhance the safety alignment of VLMs under worst-case adversarial perturbations. ADPO introduces two key components: (1) an adversarial-trained reference model that generates human-preferred responses under worst-case perturbations, and (2) an adversary-aware DPO loss that generates winner-loser pairs accounting for adversarial distortions. By combining these innovations, ADPO ensures that VLMs remain robust and reliable even in the presence of sophisticated jailbreak attacks. Extensive experiments demonstrate that ADPO outperforms baselines in terms of both safety alignment and general utility of VLMs.