ArXiv TLDR

AdversarialCoT: Single-Document Retrieval Poisoning for LLM Reasoning

🐦 Tweet
2604.12201

Hongru Song, Yu-An Liu, Ruqing Zhang, Jiafeng Guo, Maarten de Rijke + 2 more

cs.IR

TLDR

AdversarialCoT poisons LLM reasoning in RAG by injecting a single malicious document, exploiting subtle vulnerabilities to degrade accuracy.

Key contributions

  • Introduces AdversarialCoT, a query-specific attack targeting Retrieval-Augmented Generation (RAG) systems.
  • Poisons LLM reasoning by injecting only one malicious document into the retrieval corpus.
  • Extracts the target LLM's reasoning framework to guide adversarial content creation.
  • Iteratively refines the adversarial document to expose and exploit critical reasoning vulnerabilities.

Why it matters

This work exposes a significant security risk in RAG systems, demonstrating how a single poisoned document can severely degrade LLM reasoning accuracy. It offers vital insights for designing more robust and secure LLM reasoning pipelines.

Original Abstract

Retrieval-augmented generation (RAG) enhances large language model (LLM) reasoning by retrieving external documents, but also opens up new attack surfaces. We study knowledge-base poisoning attacks in RAG, where an attacker injects malicious content into the retrieval corpus, which is then naturally surfaced by the retriever and consumed by the LLM during reasoning. Unlike prior work that floods the corpus with poisoned documents, we propose AdversarialCoT, a query-specific attack that poisons only a single document in the corpus. AdversarialCoT first extracts the target LLM's reasoning framework to guide the construction of an initial adversarial chain-of-thought (CoT). The adversarial document is iteratively refined through interactions with the LLM, progressively exposing and exploiting critical reasoning vulnerabilities. Experiments on benchmark LLMs show that a single adversarial document can significantly degrade reasoning accuracy, revealing subtle yet impactful weaknesses. This study exposes security risks in RAG systems and provides actionable insights for designing more robust LLM reasoning pipelines.

📬 Weekly AI Paper Digest

Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.