ICLShield: Exploring and Mitigating In-Context Learning Backdoor Attacks

• URL: http://arxiv.org/abs/2507.01321v1
• Date: Wed, 02 Jul 2025 03:09:20 GMT
• Title: ICLShield: Exploring and Mitigating In-Context Learning Backdoor Attacks
• Authors: Zhiyao Ren, Siyuan Liang, Aishan Liu, Dacheng Tao,
• Abstract summary: In-context learning (ICL) has demonstrated remarkable success in large language models (LLMs)<n>In this paper, we propose, for the first time, the dual-learning hypothesis, which posits that LLMs simultaneously learn both the task-relevant latent concepts and backdoor latent concepts.<n>Motivated by these findings, we propose ICLShield, a defense mechanism that dynamically adjusts the concept preference ratio.
• Score: 61.06621533874629
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: In-context learning (ICL) has demonstrated remarkable success in large language models (LLMs) due to its adaptability and parameter-free nature. However, it also introduces a critical vulnerability to backdoor attacks, where adversaries can manipulate LLM behaviors by simply poisoning a few ICL demonstrations. In this paper, we propose, for the first time, the dual-learning hypothesis, which posits that LLMs simultaneously learn both the task-relevant latent concepts and backdoor latent concepts within poisoned demonstrations, jointly influencing the probability of model outputs. Through theoretical analysis, we derive an upper bound for ICL backdoor effects, revealing that the vulnerability is dominated by the concept preference ratio between the task and the backdoor. Motivated by these findings, we propose ICLShield, a defense mechanism that dynamically adjusts the concept preference ratio. Our method encourages LLMs to select clean demonstrations during the ICL phase by leveraging confidence and similarity scores, effectively mitigating susceptibility to backdoor attacks. Extensive experiments across multiple LLMs and tasks demonstrate that our method achieves state-of-the-art defense effectiveness, significantly outperforming existing approaches (+26.02% on average). Furthermore, our method exhibits exceptional adaptability and defensive performance even for closed-source models (e.g., GPT-4).

Related papers

• Satori: Reinforcement Learning with Chain-of-Action-Thought Enhances LLM Reasoning via Autoregressive Search [57.28671084993782]
  Large language models (LLMs) have demonstrated remarkable reasoning capabilities across diverse domains.<n>Recent studies have shown that increasing test-time computation enhances LLMs' reasoning capabilities.<n>We propose a two-stage training paradigm: 1) a small-scale format tuning stage to internalize the COAT reasoning format and 2) a large-scale self-improvement stage leveraging reinforcement learning.
  arXiv  Detail & Related papers  (2025-02-04T17:26:58Z)
• What Makes In-context Learning Effective for Mathematical Reasoning: A Theoretical Analysis [81.15503859645149]
  In this paper, we aim to theoretically analyze the impact of in-context demonstrations on large language models' reasoning performance.<n>We propose a straightforward, generalizable, and low-complexity demonstration selection method named LMS3.
  arXiv  Detail & Related papers  (2024-12-11T11:38:11Z)
• Cognitive Overload Attack:Prompt Injection for Long Context [39.61095361609769]
  Large Language Models (LLMs) have demonstrated remarkable capabilities in performing tasks without needing explicit retraining. This capability, known as In-Context Learning (ICL), exposes LLMs to adversarial prompts and jailbreaks that manipulate safety-trained LLMs into generating undesired or harmful output. We apply the principles of Cognitive Load Theory in LLMs and empirically validate that similar to human cognition, LLMs also suffer from cognitive overload. We show that advanced models such as GPT-4, Claude-3.5 Sonnet, Claude-3 OPUS, Llama-3-70B-Instruct, Gemini-1.0-Pro, and
  arXiv  Detail & Related papers  (2024-10-15T04:53:34Z)
• Robust LLM safeguarding via refusal feature adversarial training [15.76605079209956]
  Large language models (LLMs) are vulnerable to adversarial attacks that can elicit harmful responses.<n>We propose Refusal Feature Adrial Training (ReFAT), a novel algorithm that efficiently performs adversarial training.<n>Experiment results show that ReFAT significantly improves the robustness of three popular LLMs against a wide range of adversarial attacks.
  arXiv  Detail & Related papers  (2024-09-30T08:41:39Z)
• Evaluating and Safeguarding the Adversarial Robustness of Retrieval-Based In-Context Learning [21.018893978967053]
  In-Context Learning (ICL) is sensitive to the choice, order, and verbaliser used to encode the demonstrations in the prompt. Retrieval-Augmented ICL methods try to address this problem by leveraging retrievers to extract semantically related examples as demonstrations. Our study reveals that retrieval-augmented models can enhance robustness against test sample attacks. We introduce an effective training-free adversarial defence method, DARD, which enriches the example pool with those attacked samples.
  arXiv  Detail & Related papers  (2024-05-24T23:56:36Z)
• Data Poisoning for In-context Learning [49.77204165250528]
  In-context learning (ICL) has been recognized for its innovative ability to adapt to new tasks.<n>This paper delves into the critical issue of ICL's susceptibility to data poisoning attacks.<n>We introduce ICLPoison, a specialized attacking framework conceived to exploit the learning mechanisms of ICL.
  arXiv  Detail & Related papers  (2024-02-03T14:20:20Z)
• Benchmarking and Defending Against Indirect Prompt Injection Attacks on Large Language Models [79.0183835295533]
  We introduce the first benchmark for indirect prompt injection attacks, named BIPIA, to assess the risk of such vulnerabilities.<n>Our analysis identifies two key factors contributing to their success: LLMs' inability to distinguish between informational context and actionable instructions, and their lack of awareness in avoiding the execution of instructions within external content.<n>We propose two novel defense mechanisms-boundary awareness and explicit reminder-to address these vulnerabilities in both black-box and white-box settings.
  arXiv  Detail & Related papers  (2023-12-21T01:08:39Z)
• BadCLIP: Dual-Embedding Guided Backdoor Attack on Multimodal Contrastive Learning [85.2564206440109]
  This paper reveals the threats in this practical scenario that backdoor attacks can remain effective even after defenses. We introduce the emphtoolns attack, which is resistant to backdoor detection and model fine-tuning defenses.
  arXiv  Detail & Related papers  (2023-11-20T02:21:49Z)
• Hijacking Large Language Models via Adversarial In-Context Learning [10.416972293173993]
  In-context learning (ICL) has emerged as a powerful paradigm leveraging LLMs for specific downstream tasks by utilizing labeled examples as demonstrations (demos) in the preconditioned prompts.<n>Existing attacks are either easy to detect, require a trigger in user input, or lack specificity towards ICL.<n>This work introduces a novel transferable prompt injection attack against ICL, aiming to hijack LLMs to generate the target output or elicit harmful responses.
  arXiv  Detail & Related papers  (2023-11-16T15:01:48Z)
• Jailbreak and Guard Aligned Language Models with Only Few In-Context Demonstrations [38.437893814759086]
  Large Language Models (LLMs) have shown remarkable success in various tasks, yet their safety and the risk of generating harmful content remain pressing concerns. We propose the In-Context Attack (ICA) which employs harmful demonstrations to subvert LLMs, and the In-Context Defense (ICD) which bolsters model resilience through examples that demonstrate refusal to produce harmful responses.
  arXiv  Detail & Related papers  (2023-10-10T07:50:29Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.