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

Extensive work has been devoted to improving the safety mechanism of LargeLanguage Models (LLMs). However, in specific scenarios, LLMs still generateharmful responses when faced with malicious instructions, a phenomenon referredto as "Jailbreak Attack". In our research, we introduce a novel jailbreakattack method (\textbf{RADIAL}), which consists of two steps: 1) InherentResponse Tendency Analysis: we analyze the inherent affirmation and rejectiontendency of LLMs to react to real-world instructions. 2) Real-WorldInstructions-Driven Jailbreak: based on our analysis, we strategically chooseseveral real-world instructions and embed malicious instructions into them toamplify the LLM's potential to generate harmful responses. On three open-sourcehuman-aligned LLMs, our method achieves excellent jailbreak attack performancefor both Chinese and English malicious instructions. Besides, we guideddetailed ablation experiments and verified the effectiveness of our core idea"Inherent Response Tendency Analysis". Our exploration also exposes thevulnerability of LLMs to being induced into generating more detailed harmfulresponses in subsequent rounds of dialogue.

Analyzing the Inherent Response Tendency of LLMs: Real-World Instructions-Driven Jailbreak