Learning To See But Forgetting To Follow: Visual Instruction Tuning Makes LLMs More Prone To Jailbreak Attacks
This highlights a critical safety vulnerability in VLMs, which is an incremental but important finding for AI security and alignment research.
The paper investigates the safety of Vision-Language Models (VLMs) by showing that three state-of-the-art VLMs are more susceptible to jailbreaking attacks compared to their LLM backbones, with visual instruction-tuning causing a forgetting effect on safety guardrails.
Augmenting Large Language Models (LLMs) with image-understanding capabilities has resulted in a boom of high-performing Vision-Language models (VLMs). While studying the alignment of LLMs to human values has received widespread attention, the safety of VLMs has not received the same attention. In this paper, we explore the impact of jailbreaking on three state-of-the-art VLMs, each using a distinct modeling approach. By comparing each VLM to their respective LLM backbone, we find that each VLM is more susceptible to jailbreaking. We consider this as an undesirable outcome from visual instruction-tuning, which imposes a forgetting effect on an LLM's safety guardrails. Therefore, we provide recommendations for future work based on evaluation strategies that aim to highlight the weaknesses of a VLM, as well as take safety measures into account during visual instruction tuning.