Targeting Alignment: Extracting Safety Classifiers of Aligned LLMs

• URL: http://arxiv.org/abs/2501.16534v1
• Date: Mon, 27 Jan 2025 22:13:05 GMT
• Title: Targeting Alignment: Extracting Safety Classifiers of Aligned LLMs
• Authors: Jean-Charles Noirot Ferrand, Yohan Beugin, Eric Pauley, Ryan Sheatsley, Patrick McDaniel,
• Abstract summary: Alignment in large language models (LLMs) is used to enforce guidelines such as safety. Yet, alignment fails in the face of jailbreak attacks that modify inputs to induce unsafe outputs. We present and evaluate a method to assess the robustness of LLM alignment.
• Score: 4.492376241514766
• License:
• Abstract: Alignment in large language models (LLMs) is used to enforce guidelines such as safety. Yet, alignment fails in the face of jailbreak attacks that modify inputs to induce unsafe outputs. In this paper, we present and evaluate a method to assess the robustness of LLM alignment. We observe that alignment embeds a safety classifier in the target model that is responsible for deciding between refusal and compliance. We seek to extract an approximation of this classifier, called a surrogate classifier, from the LLM. We develop an algorithm for identifying candidate classifiers from subsets of the LLM model. We evaluate the degree to which the candidate classifiers approximate the model's embedded classifier in benign (F1 score) and adversarial (using surrogates in a white-box attack) settings. Our evaluation shows that the best candidates achieve accurate agreement (an F1 score above 80%) using as little as 20% of the model architecture. Further, we find attacks mounted on the surrogate models can be transferred with high accuracy. For example, a surrogate using only 50% of the Llama 2 model achieved an attack success rate (ASR) of 70%, a substantial improvement over attacking the LLM directly, where we only observed a 22% ASR. These results show that extracting surrogate classifiers is a viable (and highly effective) means for modeling (and therein addressing) the vulnerability of aligned models to jailbreaking attacks.

Related papers

• Palisade -- Prompt Injection Detection Framework [0.9620910657090188]
  Large Language Models are vulnerable to malicious prompt injection attacks. This paper proposes a novel NLP based approach for prompt injection detection. It emphasizes accuracy and optimization through a layered input screening process.
  arXiv  Detail & Related papers  (2024-10-28T15:47:03Z)
• Fine-tuned Large Language Models (LLMs): Improved Prompt Injection Attacks Detection [6.269725911814401]
  Large language models (LLMs) are becoming a popular tool as they have significantly advanced in their capability to tackle a wide range of language-based tasks. However, LLMs applications are highly vulnerable to prompt injection attacks, which poses a critical problem. This project explores the security vulnerabilities in relation to prompt injection attacks.
  arXiv  Detail & Related papers  (2024-10-28T00:36:21Z)
• A Realistic Threat Model for Large Language Model Jailbreaks [87.64278063236847]
  In this work, we propose a unified threat model for the principled comparison of jailbreak attacks. Our threat model combines constraints in perplexity, measuring how far a jailbreak deviates from natural text. We adapt popular attacks to this new, realistic threat model, with which we, for the first time, benchmark these attacks on equal footing.
  arXiv  Detail & Related papers  (2024-10-21T17:27:01Z)
• Model Inversion Attacks Through Target-Specific Conditional Diffusion Models [54.69008212790426]
  Model inversion attacks (MIAs) aim to reconstruct private images from a target classifier's training set, thereby raising privacy concerns in AI applications. Previous GAN-based MIAs tend to suffer from inferior generative fidelity due to GAN's inherent flaws and biased optimization within latent space. We propose Diffusion-based Model Inversion (Diff-MI) attacks to alleviate these issues.
  arXiv  Detail & Related papers  (2024-07-16T06:38:49Z)
• Cycles of Thought: Measuring LLM Confidence through Stable Explanations [53.15438489398938]
  Large language models (LLMs) can reach and even surpass human-level accuracy on a variety of benchmarks, but their overconfidence in incorrect responses is still a well-documented failure mode. We propose a framework for measuring an LLM's uncertainty with respect to the distribution of generated explanations for an answer.
  arXiv  Detail & Related papers  (2024-06-05T16:35:30Z)
• Defending Large Language Models Against Attacks With Residual Stream Activation Analysis [0.0]
  Large Language Models (LLMs) are vulnerable to adversarial threats. This paper presents an innovative defensive strategy, given white box access to an LLM. We apply a novel methodology for analyzing distinctive activation patterns in the residual streams for attack prompt classification.
  arXiv  Detail & Related papers  (2024-06-05T13:06:33Z)
• Self-Evaluation Improves Selective Generation in Large Language Models [54.003992911447696]
  We reformulate open-ended generation tasks into token-level prediction tasks. We instruct an LLM to self-evaluate its answers. We benchmark a range of scoring methods based on self-evaluation.
  arXiv  Detail & Related papers  (2023-12-14T19:09:22Z)
• Practical Membership Inference Attacks against Fine-tuned Large Language Models via Self-prompt Calibration [32.15773300068426]
  Membership Inference Attacks aim to infer whether a target data record has been utilized for model training. We propose a Membership Inference Attack based on Self-calibrated Probabilistic Variation (SPV-MIA)
  arXiv  Detail & Related papers  (2023-11-10T13:55:05Z)
• Fake Alignment: Are LLMs Really Aligned Well? [91.26543768665778]
  This study investigates the substantial discrepancy in performance between multiple-choice questions and open-ended questions. Inspired by research on jailbreak attack patterns, we argue this is caused by mismatched generalization.
  arXiv  Detail & Related papers  (2023-11-10T08:01:23Z)
• Unstoppable Attack: Label-Only Model Inversion via Conditional Diffusion Model [14.834360664780709]
  Model attacks (MIAs) aim to recover private data from inaccessible training sets of deep learning models. This paper develops a novel MIA method, leveraging a conditional diffusion model (CDM) to recover representative samples under the target label. Experimental results show that this method can generate similar and accurate samples to the target label, outperforming generators of previous approaches.
  arXiv  Detail & Related papers  (2023-07-17T12:14:24Z)
• Learning and Evaluating Representations for Deep One-class Classification [59.095144932794646]
  We present a two-stage framework for deep one-class classification. We first learn self-supervised representations from one-class data, and then build one-class classifiers on learned representations. In experiments, we demonstrate state-of-the-art performance on visual domain one-class classification benchmarks.
  arXiv  Detail & Related papers  (2020-11-04T23:33:41Z)

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.