Understanding and Defending VLM Jailbreaks via Jailbreak-Related Representation Shift
This addresses a critical safety issue for users of VLMs by providing a unified explanation and defense against jailbreaks, though it is incremental as it builds on existing understanding of representation shifts.
The paper tackles the problem of vision-language models (VLMs) being vulnerable to jailbreaks when images are added to harmful text prompts, by identifying a distinct internal representation shift caused by the visual modality that leads to failure in triggering refusal, and proposes a defense method (JRS-Rem) that removes this shift, achieving strong defense across multiple scenarios while maintaining performance on benign tasks.
Large vision-language models (VLMs) often exhibit weakened safety alignment with the integration of the visual modality. Even when text prompts contain explicit harmful intent, adding an image can substantially increase jailbreak success rates. In this paper, we observe that VLMs can clearly distinguish benign inputs from harmful ones in their representation space. Moreover, even among harmful inputs, jailbreak samples form a distinct internal state that is separable from refusal samples. These observations suggest that jailbreaks do not arise from a failure to recognize harmful intent. Instead, the visual modality shifts representations toward a specific jailbreak state, thereby leading to a failure to trigger refusal. To quantify this transition, we identify a jailbreak direction and define the jailbreak-related shift as the component of the image-induced representation shift along this direction. Our analysis shows that the jailbreak-related shift reliably characterizes jailbreak behavior, providing a unified explanation for diverse jailbreak scenarios. Finally, we propose a defense method that enhances VLM safety by removing the jailbreak-related shift (JRS-Rem) at inference time. Experiments show that JRS-Rem provides strong defense across multiple scenarios while preserving performance on benign tasks.