BadThink: Triggered Overthinking Attacks on Chain-of-Thought Reasoning in Large Language Models
Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (38) · 32141-32149
DOI: 10.1609/aaai.v40i38.40486 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Shuaitong Liu und 5 weiteren Autor:innen. Die Arbeit wurde bisher 4-mal zitiert (Crossref).
Abstract (Originalsprache)
Recent advances in Chain-of-Thought (CoT) prompting have substantially improved the reasoning capabilities of large language models (LLMs), but have also introduced their computational efficiency as a new attack surface. In this paper, we propose BadThink, the first backdoor attack designed to deliberately induce "overthinking" behavior in CoT-enabled LLMs while ensuring stealth. When activated by carefully crafted trigger prompts, BadThink manipulates the model to generate inflated reasoning traces—producing unnecessarily redundant thought processes while preserving the consistency of final outputs. This subtle attack vector creates a covert form of performance degradation that significantly increases computational costs and inference time while remaining difficult to detect through conventional output evaluation methods. We implement this attack through a sophisticated poisoning-based fine-tuning strategy, employing a novel LLM-based iterative optimization process to embed the behavior by generating highly naturalistic poisoned data. Our experiments on multiple state-of-the-art models and reasoning tasks show that BadThink consistently increases reasoning trace lengths—achieving an over 17× increase on the MATH-500 dataset—while remaining stealthy and robust. This work reveals a critical, previously unexplored vulnerability where reasoning efficiency can be covertly manipulated, demonstrating a new class of sophisticated attacks against CoT-enabled systems.
Zitieren
APA
Liu, S., Li, R., Yu, L., Zhang, L., Liu, Z., & Jin, G. (2026). BadThink: Triggered Overthinking Attacks on Chain-of-Thought Reasoning in Large Language Models. Proceedings of the AAAI Conference on Artificial Intelligence, 40(38), 32141-32149. https://doi.org/10.1609/aaai.v40i38.40486
Vancouver
Liu S, Li R, Yu L, Zhang L, Liu Z, Jin G. BadThink: Triggered Overthinking Attacks on Chain-of-Thought Reasoning in Large Language Models. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(38):32141-32149. doi:10.1609/aaai.v40i38.40486
BibTeX
@article{Liu2026,
author = {Liu, Shuaitong and Li, Renjue and Yu, Lijia and Zhang, Lijun and Liu, Zhiming and Jin, Gaojie},
title = {BadThink: Triggered Overthinking Attacks on Chain-of-Thought Reasoning in Large Language Models},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2026},
volume = {40},
number = {38},
pages = {32141--32149},
doi = {10.1609/aaai.v40i38.40486}
}
EUVIMED