Unraveling and Mitigating Safety Alignment Degradation of Vision-Language Models
This addresses a critical safety problem for users of VLMs by mitigating alignment degradation, though it is incremental as it builds on existing LLM safety methods.
The paper tackles the degradation of safety alignment in Vision-Language Models (VLMs) when integrating vision modules, showing it arises from a representation gap, and introduces Cross-Modality Representation Manipulation (CMRM) to recover safety alignment at inference time, reducing the unsafe rate of LLaVA-7B from 61.53% to 3.15% without additional training.
The safety alignment ability of Vision-Language Models (VLMs) is prone to be degraded by the integration of the vision module compared to its LLM backbone. We investigate this phenomenon, dubbed as ''safety alignment degradation'' in this paper, and show that the challenge arises from the representation gap that emerges when introducing vision modality to VLMs. In particular, we show that the representations of multi-modal inputs shift away from that of text-only inputs which represent the distribution that the LLM backbone is optimized for. At the same time, the safety alignment capabilities, initially developed within the textual embedding space, do not successfully transfer to this new multi-modal representation space. To reduce safety alignment degradation, we introduce Cross-Modality Representation Manipulation (CMRM), an inference time representation intervention method for recovering the safety alignment ability that is inherent in the LLM backbone of VLMs, while simultaneously preserving the functional capabilities of VLMs. The empirical results show that our framework significantly recovers the alignment ability that is inherited from the LLM backbone with minimal impact on the fluency and linguistic capabilities of pre-trained VLMs even without additional training. Specifically, the unsafe rate of LLaVA-7B on multi-modal input can be reduced from 61.53% to as low as 3.15% with only inference-time intervention. WARNING: This paper contains examples of toxic or harmful language.