How Few-shot Demonstrations Affect Prompt-based Defenses Against LLM Jailbreak Attacks
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866910011179925504 |
|---|---|
| author | Wang, Yanshu Yang, Shuaishuai He, Jingjing Yang, Tong |
| author_facet | Wang, Yanshu Yang, Shuaishuai He, Jingjing Yang, Tong |
| contents | Large Language Models (LLMs) face increasing threats from jailbreak attacks that bypass safety alignment. While prompt-based defenses such as Role-Oriented Prompts (RoP) and Task-Oriented Prompts (ToP) have shown effectiveness, the role of few-shot demonstrations in these defense strategies remains unclear. Prior work suggests that few-shot examples may compromise safety, but lacks investigation into how few-shot interacts with different system prompt strategies. In this paper, we conduct a comprehensive evaluation on multiple mainstream LLMs across four safety benchmarks (AdvBench, HarmBench, SG-Bench, XSTest) using six jailbreak attack methods. Our key finding reveals that few-shot demonstrations produce opposite effects on RoP and ToP: few-shot enhances RoP's safety rate by up to 4.5% through reinforcing role identity, while it degrades ToP's effectiveness by up to 21.2% through distracting attention from task instructions. Based on these findings, we provide practical recommendations for deploying prompt-based defenses in real-world LLM applications. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_04294 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | How Few-shot Demonstrations Affect Prompt-based Defenses Against LLM Jailbreak Attacks Wang, Yanshu Yang, Shuaishuai He, Jingjing Yang, Tong Computation and Language Artificial Intelligence Cryptography and Security 68T50 I.2.7; K.4.1 Large Language Models (LLMs) face increasing threats from jailbreak attacks that bypass safety alignment. While prompt-based defenses such as Role-Oriented Prompts (RoP) and Task-Oriented Prompts (ToP) have shown effectiveness, the role of few-shot demonstrations in these defense strategies remains unclear. Prior work suggests that few-shot examples may compromise safety, but lacks investigation into how few-shot interacts with different system prompt strategies. In this paper, we conduct a comprehensive evaluation on multiple mainstream LLMs across four safety benchmarks (AdvBench, HarmBench, SG-Bench, XSTest) using six jailbreak attack methods. Our key finding reveals that few-shot demonstrations produce opposite effects on RoP and ToP: few-shot enhances RoP's safety rate by up to 4.5% through reinforcing role identity, while it degrades ToP's effectiveness by up to 21.2% through distracting attention from task instructions. Based on these findings, we provide practical recommendations for deploying prompt-based defenses in real-world LLM applications. |
| title | How Few-shot Demonstrations Affect Prompt-based Defenses Against LLM Jailbreak Attacks |
| topic | Computation and Language Artificial Intelligence Cryptography and Security 68T50 I.2.7; K.4.1 |
| url | https://arxiv.org/abs/2602.04294 |