Breaking the Loop: Detecting and Mitigating Denial-of-Service Vulnerabilities in Large Language Models

2025-03-01

AI Summary

This work identifies a critical denial-of-service (DoS) vulnerability in large language models (LLMs), wherein repetitive generation of similar or identical outputs induces severe latency spikes. To address this, we propose RecurrentGenerator, the first black-box framework for discovering cyclic generation vulnerabilities via evolutionary search to automatically trigger recurrent behavior. We further design RecurrentDetector, a lightweight real-time detector that trains a binary classifier on neural activation patterns to achieve high-accuracy, low-overhead anomaly identification. Empirical evaluation on Llama-3 and GPT-4o confirms multiple classes of cyclic generation vulnerabilities. RecurrentDetector achieves 95.24% accuracy and an F1 score of 0.87. This is the first systematic study to expose the DoS risks inherent in LLM recurrent generation. To foster community defense research, we open-source both our detection tools and benchmark dataset.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large Language Models (LLMs) have significantly advanced text understanding and generation, becoming integral to applications across education, software development, healthcare, entertainment, and legal services. Despite considerable progress in improving model reliability, latency remains under-explored, particularly through recurrent generation, where models repeatedly produce similar or identical outputs, causing increased latency and potential Denial-of-Service (DoS) vulnerabilities. We propose RecurrentGenerator, a black-box evolutionary algorithm that efficiently identifies recurrent generation scenarios in prominent LLMs like LLama-3 and GPT-4o. Additionally, we introduce RecurrentDetector, a lightweight real-time classifier trained on activation patterns, achieving 95.24% accuracy and an F1 score of 0.87 in detecting recurrent loops. Our methods provide practical solutions to mitigate latency-related vulnerabilities, and we publicly share our tools and data to support further research.

Problem

• Detect and mitigate Denial-of-Service vulnerabilities in LLMs
• Address latency issues from recurrent generation in LLMs
• Develop tools to identify and prevent recurrent output loops

Innovation

• Black-box evolutionary algorithm for recurrent generation detection
• Lightweight real-time classifier with high accuracy
• Public tools and data for vulnerability mitigation