Using Mechanistic Interpretability to Craft Adversarial Attacks against Large Language Models

• URL: http://arxiv.org/abs/2503.06269v1
• Date: Sat, 08 Mar 2025 16:29:45 GMT
• Title: Using Mechanistic Interpretability to Craft Adversarial Attacks against Large Language Models
• Authors: Thomas Winninger, Boussad Addad, Katarzyna Kapusta,
• Abstract summary: We introduce a novel white-box approach for creating adversarial perturbations against LLMs.<n>We first identify acceptance subspaces - sets of feature vectors that do not trigger the model's refusal mechanisms.<n>We then use gradient-based optimization to reroute embeddings from refusal subspaces to acceptance subspaces, effectively achieving jailbreaks.
• Score: 1.6112718683989882
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Traditional white-box methods for creating adversarial perturbations against LLMs typically rely only on gradient computation from the targeted model, ignoring the internal mechanisms responsible for attack success or failure. Conversely, interpretability studies that analyze these internal mechanisms lack practical applications beyond runtime interventions. We bridge this gap by introducing a novel white-box approach that leverages mechanistic interpretability techniques to craft practical adversarial inputs. Specifically, we first identify acceptance subspaces - sets of feature vectors that do not trigger the model's refusal mechanisms - then use gradient-based optimization to reroute embeddings from refusal subspaces to acceptance subspaces, effectively achieving jailbreaks. This targeted approach significantly reduces computation cost, achieving attack success rates of 80-95\% on state-of-the-art models including Gemma2, Llama3.2, and Qwen2.5 within minutes or even seconds, compared to existing techniques that often fail or require hours of computation. We believe this approach opens a new direction for both attack research and defense development. Furthermore, it showcases a practical application of mechanistic interpretability where other methods are less efficient, which highlights its utility. The code and generated datasets are available at https://github.com/Sckathach/subspace-rerouting.

Related papers

• STShield: Single-Token Sentinel for Real-Time Jailbreak Detection in Large Language Models [31.35788474507371]
  Large Language Models (LLMs) have become increasingly vulnerable to jailbreak attacks. We present STShield, a lightweight framework for real-time jailbroken judgement.
  arXiv  Detail & Related papers  (2025-03-23T04:23:07Z)
• No Prior Mask: Eliminate Redundant Action for Deep Reinforcement Learning [13.341525656639583]
  Large action space is one fundamental obstacle to deploying Reinforcement Learning methods in the real world. We propose a novel redundant action filtering mechanism named No Prior Mask (NPM)
  arXiv  Detail & Related papers  (2023-12-11T09:56:02Z)
• Rethinking PGD Attack: Is Sign Function Necessary? [131.6894310945647]
  We present a theoretical analysis of how such sign-based update algorithm influences step-wise attack performance. We propose a new raw gradient descent (RGD) algorithm that eliminates the use of sign. The effectiveness of the proposed RGD algorithm has been demonstrated extensively in experiments.
  arXiv  Detail & Related papers  (2023-12-03T02:26:58Z)
• Defense Against Model Extraction Attacks on Recommender Systems [53.127820987326295]
  We introduce Gradient-based Ranking Optimization (GRO) to defend against model extraction attacks on recommender systems. GRO aims to minimize the loss of the protected target model while maximizing the loss of the attacker's surrogate model. Results show GRO's superior effectiveness in defending against model extraction attacks.
  arXiv  Detail & Related papers  (2023-10-25T03:30:42Z)
• Open Sesame! Universal Black Box Jailbreaking of Large Language Models [0.0]
  Large language models (LLMs) are designed to provide helpful and safe responses. LLMs often rely on alignment techniques to align with user intent and social guidelines. We introduce a novel approach that employs a genetic algorithm (GA) to manipulate LLMs when model architecture and parameters are inaccessible.
  arXiv  Detail & Related papers  (2023-09-04T08:54:20Z)
• RamBoAttack: A Robust Query Efficient Deep Neural Network Decision Exploit [9.93052896330371]
  We develop a robust query efficient attack capable of avoiding entrapment in a local minimum and misdirection from noisy gradients. The RamBoAttack is more robust to the different sample inputs available to an adversary and the targeted class.
  arXiv  Detail & Related papers  (2021-12-10T01:25:24Z)
• Transferable Sparse Adversarial Attack [62.134905824604104]
  We introduce a generator architecture to alleviate the overfitting issue and thus efficiently craft transferable sparse adversarial examples. Our method achieves superior inference speed, 700$times$ faster than other optimization-based methods.
  arXiv  Detail & Related papers  (2021-05-31T06:44:58Z)
• Automated Decision-based Adversarial Attacks [48.01183253407982]
  We consider the practical and challenging decision-based black-box adversarial setting. Under this setting, the attacker can only acquire the final classification labels by querying the target model. We propose to automatically discover decision-based adversarial attack algorithms.
  arXiv  Detail & Related papers  (2021-05-09T13:15:10Z)
• Adversarial EXEmples: A Survey and Experimental Evaluation of Practical Attacks on Machine Learning for Windows Malware Detection [67.53296659361598]
  adversarial EXEmples can bypass machine learning-based detection by perturbing relatively few input bytes. We develop a unifying framework that does not only encompass and generalize previous attacks against machine-learning models, but also includes three novel attacks. These attacks, named Full DOS, Extend and Shift, inject the adversarial payload by respectively manipulating the DOS header, extending it, and shifting the content of the first section.
  arXiv  Detail & Related papers  (2020-08-17T07:16:57Z)
• A Self-supervised Approach for Adversarial Robustness [105.88250594033053]
  Adversarial examples can cause catastrophic mistakes in Deep Neural Network (DNNs) based vision systems. This paper proposes a self-supervised adversarial training mechanism in the input space. It provides significant robustness against the textbfunseen adversarial attacks.
  arXiv  Detail & Related papers  (2020-06-08T20:42:39Z)
• Luring of transferable adversarial perturbations in the black-box paradigm [0.0]
  We present a new approach to improve the robustness of a model against black-box transfer attacks. A removable additional neural network is included in the target model, and is designed to induce the textitluring effect. Our deception-based method only needs to have access to the predictions of the target model and does not require a labeled data set.
  arXiv  Detail & Related papers  (2020-04-10T06:48:36Z)

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.