Token-Efficient Prompt Injection Attack: Provoking Cessation in LLM Reasoning via Adaptive Token Compression

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cui, Yu, Cai, Yujun, Wang, Yiwei
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912353190150144
author Cui, Yu
Cai, Yujun
Wang, Yiwei
author_facet Cui, Yu
Cai, Yujun
Wang, Yiwei
contents While reasoning large language models (LLMs) demonstrate remarkable performance across various tasks, they also contain notable security vulnerabilities. Recent research has uncovered a "thinking-stopped" vulnerability in DeepSeek-R1, where model-generated reasoning tokens can forcibly interrupt the inference process, resulting in empty responses that compromise LLM-integrated applications. However, existing methods triggering this vulnerability require complex mathematical word problems with long prompts--even exceeding 5,000 tokens. To reduce the token cost and formally define this vulnerability, we propose a novel prompt injection attack named "Reasoning Interruption Attack", based on adaptive token compression. We demonstrate that simple standalone arithmetic tasks can effectively trigger this vulnerability, and the prompts based on such tasks exhibit simpler logical structures than mathematical word problems. We develop a systematic approach to efficiently collect attack prompts and an adaptive token compression framework that utilizes LLMs to automatically compress these prompts. Experiments show our compression framework significantly reduces prompt length while maintaining effective attack capabilities. We further investigate the attack's performance via output prefix and analyze the underlying causes of the vulnerability, providing valuable insights for improving security in reasoning LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2504_20493
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Token-Efficient Prompt Injection Attack: Provoking Cessation in LLM Reasoning via Adaptive Token Compression
Cui, Yu
Cai, Yujun
Wang, Yiwei
Cryptography and Security
Artificial Intelligence
While reasoning large language models (LLMs) demonstrate remarkable performance across various tasks, they also contain notable security vulnerabilities. Recent research has uncovered a "thinking-stopped" vulnerability in DeepSeek-R1, where model-generated reasoning tokens can forcibly interrupt the inference process, resulting in empty responses that compromise LLM-integrated applications. However, existing methods triggering this vulnerability require complex mathematical word problems with long prompts--even exceeding 5,000 tokens. To reduce the token cost and formally define this vulnerability, we propose a novel prompt injection attack named "Reasoning Interruption Attack", based on adaptive token compression. We demonstrate that simple standalone arithmetic tasks can effectively trigger this vulnerability, and the prompts based on such tasks exhibit simpler logical structures than mathematical word problems. We develop a systematic approach to efficiently collect attack prompts and an adaptive token compression framework that utilizes LLMs to automatically compress these prompts. Experiments show our compression framework significantly reduces prompt length while maintaining effective attack capabilities. We further investigate the attack's performance via output prefix and analyze the underlying causes of the vulnerability, providing valuable insights for improving security in reasoning LLMs.
title Token-Efficient Prompt Injection Attack: Provoking Cessation in LLM Reasoning via Adaptive Token Compression
topic Cryptography and Security
Artificial Intelligence
url https://arxiv.org/abs/2504.20493