Overthinking Large Language Models: New DoS Threat to Reasoning Models Unveiled

The paper introduces a black‑box hierarchical genetic algorithm that automatically perturbs the logical structure of reasoning questions to induce excessive chain‑of‑thought in large language models, dramatically inflating output tokens (up to 26.1× on MATH) and creating a DoS‑style resource‑exhaustion attack, with extensive experiments across multiple models demonstrating the vulnerability and its transferability.

DoS attackLLMhierarchical genetic algorithm

0 likes · 10 min read

Overthinking Large Language Models: New DoS Threat to Reasoning Models Unveiled