When Smiley Turns Hostile: Interpreting How Emojis Trigger LLMs’ Toxicity
Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (36) · 30431-30439
DOI: 10.1609/aaai.v40i36.40296 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Shiyao Cui und 8 weiteren Autor:innen.
Abstract (Originalsprache)
Emojis are globally used non-verbal cues in digital communication, and extensive research has examined how large language models (LLMs) understand and utilize emojis across contexts. While usually associated with friendliness or playfulness, it is observed that emojis may trigger toxic content generation in LLMs. Motivated by such a observation, we aim to investigate: (1) whether emojis can clearly enhance the toxicity generation in LLMs and (2) how to interpret this phenomenon.* We begin with a comprehensive exploration of emoji-triggered LLM toxicity generation by automating the construction of prompts with emojis to subtly express toxic intent. Experiments across 5 mainstream languages on 7 famous LLMs along with jailbreak tasks demonstrate that prompts with emojis could easily induce toxicity generation. To understand this phenomenon, we conduct model-level interpretations spanning semantic cognition, sequence generation and tokenization, suggesting that emojis can act as a heterogeneous semantic channel to bypass the safety mechanisms. To pursue deeper insights, we further probe the pre-training corpus and uncover potential correlation between the emoji-related data polution with the toxicity generation behaviors.
Zitieren
APA
Cui, S., Feng, X., Wang, Y., Yang, J., Zhang, Z., Sikdar, B., Wang, H., Qiu, H., & Huang, M. (2026). When Smiley Turns Hostile: Interpreting How Emojis Trigger LLMs’ Toxicity. Proceedings of the AAAI Conference on Artificial Intelligence, 40(36), 30431-30439. https://doi.org/10.1609/aaai.v40i36.40296
Vancouver
Cui S, Feng X, Wang Y, Yang J, Zhang Z, Sikdar B, et al. When Smiley Turns Hostile: Interpreting How Emojis Trigger LLMs’ Toxicity. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(36):30431-30439. doi:10.1609/aaai.v40i36.40296
BibTeX
@article{Cui2026,
author = {Cui, Shiyao and Feng, Xijia and Wang, Yingkang and Yang, Junxiao and Zhang, Zhexin and Sikdar, Biplab and Wang, Hongning and Qiu, Han and Huang, Minlie},
title = {When Smiley Turns Hostile: Interpreting How Emojis Trigger LLMs’ Toxicity},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2026},
volume = {40},
number = {36},
pages = {30431--30439},
doi = {10.1609/aaai.v40i36.40296}
}
EUVIMED