EUVIMEDEUVIMED entdecken
Crossref

Human-Readable Adversarial Prompts: An Investigation into LLM Vulnerabilities Using Situational Context

Nilanjana Das, Edward Raff, Aman Chadha, Manas Gaur

ACM Transactions on Intelligent Systems and Technology · 07.05.2026

DOI: 10.1145/3815174 ↗

Kurz gefasst

Fachartikel in der Zeitschrift ACM Transactions on Intelligent Systems and Technology, online erschienen am 07.05.2026. Verfasst von Nilanjana Das und 3 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 31 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Large Language Models (LLMs) remain vulnerable to adversarial prompts that elicit harmful responses. While many existing safety systems are better able to detect overtly nonsensical attack strings, human-readable prompts embedded in plausible situational contexts remain harder to identify and evaluate. This paper presents an empirical investigation of human-readable, situation-driven adversarial prompts for assessing LLM robustness. First, we use movie scripts as situational contexts (e.g., crime narratives) to construct natural-looking prompts that bypass safety mechanisms. Second, we transform adversarial gibberish into coherent, innocuous-appearing text that retains exploitation capability within these contextual frameworks. Third, we enhance the AdvPrompter framework with p-nucleus sampling to generate diverse human-readable attacks, substantially improving success rates against models including GPT-3.5 and Gemma-7b. We validate our approach through multi-method evaluation: automated harmfulness scoring (GPT-4o-mini), independent human assessment by 10 raters across 80 samples, and Elo rating meta-analysis of judge reliability. These findings highlight the need for safety mechanisms that can detect not only nonsensical jailbreak strings, but also coherent adversarial content embedded in realistic narrative contexts. The code for this study is publicly available. 1

Zitieren

APA

Das, N., Raff, E., Chadha, A., & Gaur, M. (2026). Human-Readable Adversarial Prompts: An Investigation into LLM Vulnerabilities Using Situational Context. ACM Transactions on Intelligent Systems and Technology. https://doi.org/10.1145/3815174

Vancouver

Das N, Raff E, Chadha A, Gaur M. Human-Readable Adversarial Prompts: An Investigation into LLM Vulnerabilities Using Situational Context. ACM Transactions on Intelligent Systems and Technology. 2026. doi:10.1145/3815174

BibTeX
@article{Das2026,
  author = {Das, Nilanjana and Raff, Edward and Chadha, Aman and Gaur, Manas},
  title = {Human-Readable Adversarial Prompts: An Investigation into LLM Vulnerabilities Using Situational Context},
  journal = {ACM Transactions on Intelligent Systems and Technology},
  year = {2026},
  doi = {10.1145/3815174}
}
Kostenlos starten →