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

Large Language Models (LLMs) are swiftly advancing in architecture andcapability, and as they integrate more deeply into complex systems, the urgencyto scrutinize their security properties grows. This paper surveys research inthe emerging interdisciplinary field of adversarial attacks on LLMs, a subfieldof trustworthy ML, combining the perspectives of Natural Language Processingand Security. Prior work has shown that even safety-aligned LLMs (viainstruction tuning and reinforcement learning through human feedback) can besusceptible to adversarial attacks, which exploit weaknesses and mislead AIsystems, as evidenced by the prevalence of `jailbreak' attacks on models likeChatGPT and Bard. In this survey, we first provide an overview of largelanguage models, describe their safety alignment, and categorize existingresearch based on various learning structures: textual-only attacks,multi-modal attacks, and additional attack methods specifically targetingcomplex systems, such as federated learning or multi-agent systems. We alsooffer comprehensive remarks on works that focus on the fundamental sources ofvulnerabilities and potential defenses. To make this field more accessible tonewcomers, we present a systematic review of existing works, a structuredtypology of adversarial attack concepts, and additional resources, includingslides for presentations on related topics at the 62nd Annual Meeting of theAssociation for Computational Linguistics (ACL'24).

Survey of Vulnerabilities in Large Language Models Revealed by Adversarial Attacks