Failures to Find Transferable Image Jailbreaks Between Vision-Language Models

The integration of new modalities into frontier AI systems increases the possibility such systems can be adversarially manipulated in undesirable ways. In this work, we focus on a popular class of vision-language models (VLMs) that generate text conditioned on visual and textual inputs. We conducted a large-scale empirical study to assess the transferability of gradient-based universal image "jailbreaks" using a diverse set of over 40 open-parameter VLMs, including 18 new VLMs that we publicly release. We find that transferable gradient-based image jailbreaks are extremely difficult to obtain. When an image jailbreak is optimized against a single VLM or against an ensemble of VLMs, the image successfully jailbreaks the attacked VLM(s), but exhibits little-to-no transfer to any other VLMs; transfer is not affected by whether the attacked and target VLMs possess matching vision backbones or language models, whether the language model underwent instructionfollowing and/or safety-alignment training, or other factors. Only two settings display partial transfer: between identically-pretrained and identically-initialized VLMs with slightly different VLM training data, and between different training checkpoints of a single VLM. Leveraging these results, we demonstrate that transfer can be significantly improved against a specific target VLM by attacking larger ensembles of "highly-similar" VLMs. These results stand in stark contrast to existing evidence of universal and transferable text jailbreaks against language models and transferable adversarial attacks against image classifiers, suggesting that VLMs may be more robust to gradient-based transfer attacks.