Adversarial Tokenization

Renato Geh, Zilei Shao, Guy Van den Broeck

Keywords: adversarial attacks, tokenization

TL;DR: We show a previously unknown vulnerability of LLMs in addressing tokenization attacks whereby simply retokenizing an unsafe request elicits dangerous responses in state-of-the-art LLMs.

Abstract: Current LLM pipelines account for only one possible tokenization for a given string, ignoring exponentially many alternative tokenizations during training and inference. For example, the standard Llama3 tokenization of penguin is [p,enguin], yet [peng,uin] is another perfectly valid alternative. In this paper, we show that despite LLMs being trained solely on one tokenization, they still retain semantic understanding of other tokenizations, raising questions about their implications in LLM safety. Put succinctly, we answer the following question: can we adversarially tokenize an obviously malicious string to evade safety and alignment restrictions? We show that not only is adversarial tokenization an effective yet previously neglected axis of attack, but it is also competitive against existing state-of-the-art adversarial approaches without changing the text of the harmful request. We empirically validate this exploit across three state-of-the-art LLMs and adversarial datasets, revealing a previously unknown vulnerability in subword models.

Archival Status: Non-archival

Paper Length: Long Paper (up to 8 pages of content)

Submission Number: 198