Small Symbols, Big Risks: Exploring Emoticon Semantic Confusion in Large Language Models
This work addresses a safety problem for users and developers of LLMs by revealing a pervasive and transferable vulnerability with potential security risks, though it is incremental in exploring a specific, previously underexplored aspect of LLM safety.
The paper identified emoticon semantic confusion as a vulnerability in Large Language Models (LLMs), where misinterpretation of ASCII-based emoticons leads to unintended actions, with an average confusion ratio exceeding 38% and over 90% of confused responses causing 'silent failures' that could result in destructive security consequences.
Emoticons are widely used in digital communication to convey affective intent, yet their safety implications for Large Language Models (LLMs) remain largely unexplored. In this paper, we identify emoticon semantic confusion, a vulnerability where LLMs misinterpret ASCII-based emoticons to perform unintended and even destructive actions. To systematically study this phenomenon, we develop an automated data generation pipeline and construct a dataset containing 3,757 code-oriented test cases spanning 21 meta-scenarios, four programming languages, and varying contextual complexities. Our study on six LLMs reveals that emoticon semantic confusion is pervasive, with an average confusion ratio exceeding 38%. More critically, over 90% of confused responses yield 'silent failures', which are syntactically valid outputs but deviate from user intent, potentially leading to destructive security consequences. Furthermore, we observe that this vulnerability readily transfers to popular agent frameworks, while existing prompt-based mitigations remain largely ineffective. We call on the community to recognize this emerging vulnerability and develop effective mitigation methods to uphold the safety and reliability of the LLM system.