This research identifies critical denial-of-service vulnerabilities in LLMs caused by recurrent generation patterns that significantly increase latency and system resource consumption.

• Reveals how malicious prompts can force LLMs into repetitive output loops
• Proposes novel detection mechanisms to identify potential DoS attacks
• Presents effective mitigation strategies to maintain system availability
• Demonstrates practical security improvements across multiple LLM architectures

As LLMs become integral to critical applications in healthcare, legal services, and software development, these security enhancements help protect against availability-based attacks that could compromise essential AI systems.

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