Why Do Aligned LLMs Remain Jailbreakable: Refusal-Escape Directions, Operator-Level Sources, and Safety-Utility Trade-off

• URL: http://arxiv.org/abs/2605.08878v1
• Date: Sat, 09 May 2026 10:57:36 GMT
• Title: Why Do Aligned LLMs Remain Jailbreakable: Refusal-Escape Directions, Operator-Level Sources, and Safety-Utility Trade-off
• Abstract summary: Aligned large language models (LLMs) remain vulnerable to jailbreak attacks.<n>We show that aligned models can still exhibit Refusal-Escape Directions (RED)<n>We then prove that RED can be exactly decomposed into contributions from operator-level sources across the model's operator structure.
• Score: 10.459125169096696
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Aligned large language models (LLMs) remain vulnerable to jailbreak attacks. Recent mechanistic studies have identified latent features and representation shifts associated with jailbreak success, but they leave a more fundamental question open: why do aligned LLMs remain jailbreakable, and what structural vulnerabilities in the model make this possible? We study this question through a continuous input-transformation view. Our theoretical finding is that aligned models can still exhibit Refusal-Escape Directions (RED): local perturbation directions around a harmful input that shift the model's behavior from refusal to answering while preserving the model's harmful-semantics interpretation. From this perspective, a jailbreak is not only a successful discrete prompt construction, but can also be understood as a refusal-to-answer behavior transition induced by continuously perturbing a harmful input along RED. We then prove that RED can be exactly decomposed into contributions from operator-level sources across the model's operator structure, and identify normalization, residual-wiring, and terminal sources as analytically constrained operator-level sources. To eliminate RED, the shared expressive modules -- self-attention and MLP -- must eliminate the contributions from these analytically constrained sources while preserving the mechanisms that support benign responses. These competing requirements give rise to a conditional safety-utility trade-off. Experiments across multiple models and attack methods empirically analyze RED from two complementary perspectives and show that added token dimensions can expose RED, while successful jailbreaks exhibit refusal-to-answer shifts largely aligned with terminal-source contributions.

Related papers

• Knowing When to Quit: Diagnosing and Training LLMs to Abort Futile Reasoning [109.33717747526583]
  Large language models generate computationally expensive yet semantically void reasoning on beyond-capability tasks.<n>We introduce textbfCaRL (textbfCapability-textbfaligned textbfReinforcement textbfL) to align behavior with capability boundaries.<n>Experiments demonstrate a substantial reduction in futile reasoning while preserving performance across task difficulties.
  arXiv  Detail & Related papers  (2026-07-31T09:30:33Z)
• Beyond the Prompt: Jailbreaking Function-Calling LLMs via Simulated Moderation Traces [32.850551109913845]
  Jailbreak attacks remain a critical threat to the safe deployment of large language models (LLMs)<n>We show that this prompt-centric paradigm overlooks a structural vulnerability in stateful, function-calling environments.<n>We exploit this architectural flaw through SMT, a black-box attack framework based on Simulated Moderation Traces.
  arXiv  Detail & Related papers  (2026-07-01T06:08:07Z)
• Benign Inputs, Harmful Outputs: Cross-Modal Jailbreaking via Distributed Semantic Recomposition [10.177453923096754]
  Multimodal Large Language Models (MLLMs) have recently demonstrated remarkable capabilities in content synthesis and autonomous reasoning.<n>Previous safety guardrails are primarily designed for unimodal textual input interception.<n>We propose Distributed Semantic Recomposition (DSR), a novel cross-modal jailbreak framework.
  arXiv  Detail & Related papers  (2026-06-01T07:49:12Z)
• SAGE: Shaping Anchors for Guided Exploration in RLVR of LLMs [55.46289074417954]
  reinforcement learning with verifiable rewards (RLVR) reliably improves pass@1 on reasoning tasks, yet often fails to yield comparable gains in pass@k.<n>We argue that a central structural constraint arises from reverse-KL regularization, which stabilizes training but inherently anchors the policy to the reference distribution.<n>We propose SAGE, a principled framework that enables controllable empirical support expansion by reshaping the reverse-KL anchor distribution itself.
  arXiv  Detail & Related papers  (2026-05-15T07:42:21Z)
• Fail-Closed Alignment for Large Language Models [4.205036273334146]
  We propose fail-closed alignment as a design principle for robust large language model safety.<n>We present a progressive alignment framework that iteratively identifies and ablates previously learned refusal directions.<n>Our mechanistic analyses confirm that models trained with our method encode multiple, causally independent refusal directions that prompt-based jailbreaks cannot suppress simultaneously.
  arXiv  Detail & Related papers  (2026-02-19T00:33:35Z)
• Jailbreaking Leaves a Trace: Understanding and Detecting Jailbreak Attacks from Internal Representations of Large Language Models [2.6140509675507384]
  We study jailbreaking from both security and interpretability perspectives.<n>We propose a tensor-based latent representation framework that captures structure in hidden activations.<n>Our results provide evidence that jailbreak behavior is rooted in identifiable internal structures.
  arXiv  Detail & Related papers  (2026-02-12T02:43:17Z)
• A Causal Perspective for Enhancing Jailbreak Attack and Defense [29.669194815878768]
  We propose a framework that integrates large language models into data-driven causal discovery.<n>We introduce a comprehensive dataset comprising 35k jailbreak attempts across seven language models.<n>Our analysis reveals that specific features, such as "Positive Character" and "Number of Task Steps", act as direct causal drivers of jailbreaks.
  arXiv  Detail & Related papers  (2026-01-31T15:20:13Z)
• Is Reasoning All You Need? Probing Bias in the Age of Reasoning Language Models [0.0]
  Reasoning Language Models (RLMs) have gained traction for their ability to perform complex, multi-step reasoning tasks.<n>While these capabilities promise improved reliability, their impact on robustness to social biases remains unclear.<n>We leverage the CLEAR-Bias benchmark to investigate the adversarial robustness of RLMs to bias elicitation.
  arXiv  Detail & Related papers  (2025-07-03T17:01:53Z)
• Cannot See the Forest for the Trees: Invoking Heuristics and Biases to Elicit Irrational Choices of LLMs [83.11815479874447]
  We propose a novel jailbreak attack framework, inspired by cognitive decomposition and biases in human cognition.<n>We employ cognitive decomposition to reduce the complexity of malicious prompts and relevance bias to reorganize prompts.<n>We also introduce a ranking-based harmfulness evaluation metric that surpasses the traditional binary success-or-failure paradigm.
  arXiv  Detail & Related papers  (2025-05-03T05:28:11Z)
• MIRAGE: Multimodal Immersive Reasoning and Guided Exploration for Red-Team Jailbreak Attacks [85.3303135160762]
  MIRAGE is a novel framework that exploits narrative-driven context and role immersion to circumvent safety mechanisms in Multimodal Large Language Models.<n>It achieves state-of-the-art performance, improving attack success rates by up to 17.5% over the best baselines.<n>We demonstrate that role immersion and structured semantic reconstruction can activate inherent model biases, facilitating the model's spontaneous violation of ethical safeguards.
  arXiv  Detail & Related papers  (2025-03-24T20:38:42Z)
• Improving LLM Safety Alignment with Dual-Objective Optimization [81.98466438000086]
  Existing training-time safety alignment techniques for large language models (LLMs) remain vulnerable to jailbreak attacks.<n>We propose an improved safety alignment that disentangles DPO objectives into two components: (1) robust refusal training, which encourages refusal even when partial unsafe generations are produced, and (2) targeted unlearning of harmful knowledge.
  arXiv  Detail & Related papers  (2025-03-05T18:01:05Z)
• JailbreakLens: Interpreting Jailbreak Mechanism in the Lens of Representation and Circuit [12.392258585661446]
  Large language models (LLMs) are vulnerable to jailbreak attacks, wherein adversarial prompts are crafted to bypass their security mechanisms and elicit unexpected responses.<n>This paper proposes JailbreakLens, an interpretation framework that analyzes jailbreak mechanisms from both representation and circuit perspectives.
  arXiv  Detail & Related papers  (2024-11-17T16:08:34Z)
• AutoDAN: Interpretable Gradient-Based Adversarial Attacks on Large Language Models [55.748851471119906]
  Safety alignment of Large Language Models (LLMs) can be compromised with manual jailbreak attacks and (automatic) adversarial attacks. Recent studies suggest that defending against these attacks is possible: adversarial attacks generate unlimited but unreadable gibberish prompts, detectable by perplexity-based filters. We introduce AutoDAN, an interpretable, gradient-based adversarial attack that merges the strengths of both attack types.
  arXiv  Detail & Related papers  (2023-10-23T17:46:07Z)

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.