English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Ensuring the safety of artificial intelligence-generated content (AIGC) is alongstanding topic in the artificial intelligence (AI) community, and thesafety concerns associated with Large Language Models (LLMs) have been widelyinvestigated. Recently, large vision-language models (VLMs) represent anunprecedented revolution, as they are built upon LLMs but can incorporateadditional modalities (e.g., images). However, the safety of VLMs lackssystematic evaluation, and there may be an overconfidence in the safetyguarantees provided by their underlying LLMs. In this paper, to demonstratethat introducing additional modality modules leads to unforeseen AI safetyissues, we propose FigStep, a straightforward yet effective jailbreakingalgorithm against VLMs. Instead of feeding textual harmful instructionsdirectly, FigStep converts the harmful content into images through typographyto bypass the safety alignment within the textual module of the VLMs, inducingVLMs to output unsafe responses that violate common AI safety policies. In ourevaluation, we manually review 46,500 model responses generated by 3 familiesof the promising open-source VLMs, i.e., LLaVA, MiniGPT4, and CogVLM (a totalof 6 VLMs). The experimental results show that FigStep can achieve an averageattack success rate of 82.50% on 500 harmful queries in 10 topics. Moreover, wedemonstrate that the methodology of FigStep can even jailbreak GPT-4V, whichalready leverages an OCR detector to filter harmful queries. Above all, ourwork reveals that VLMs are vulnerable to jailbreaking attacks, which highlightsthe necessity of novel safety alignments between visual and textual modalities.

FigStep: Jailbreaking Large Vision-language Models via Typographic Visual Prompts