Efficient Refusal Ablation in LLM through Optimal Transport

• URL: http://arxiv.org/abs/2603.04355v1
• Date: Wed, 04 Mar 2026 18:19:50 GMT
• Title: Efficient Refusal Ablation in LLM through Optimal Transport
• Authors: Geraldin Nanfack, Eugene Belilovsky, Elvis Dohmatob,
• Abstract summary: Safety-aligned language models refuse harmful requests through learned refusal behaviors encoded in their internal representations.<n>Recent activation-based jailbreaking methods circumvent these safety mechanisms by applying projections to remove refusal directions.<n>We introduce a principled framework based on optimal transport theory that transforms the entire distribution of harmful activations to match harmless ones.
• Score: 30.0180859405821
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Safety-aligned language models refuse harmful requests through learned refusal behaviors encoded in their internal representations. Recent activation-based jailbreaking methods circumvent these safety mechanisms by applying orthogonal projections to remove refusal directions, but these approaches treat refusal as a one-dimensional phenomenon and ignore the rich distributional structure of model activations. We introduce a principled framework based on optimal transport theory that transforms the entire distribution of harmful activations to match harmless ones. By combining PCA with closed-form Gaussian optimal transport, we achieve efficient computation in high-dimensional representation spaces while preserving essential geometric structure. Across six models (Llama-2, Llama-3.1, Qwen-2.5; 7B-32B parameters), our method achieves up to 11% higher attack success rates than state-of-the-art baselines while maintaining comparable perplexity, demonstrating superior preservation of model capabilities. Critically, we discover that layer-selective intervention (applying optimal transport to 1-2 carefully chosen layers at approximately 40-60% network depth) substantially outperforms full-network interventions, revealing that refusal mechanisms may be localized rather than distributed. Our analysis provides new insights into the geometric structure of safety representations and suggests that current alignment methods may be vulnerable to distributional attacks beyond simple direction removal.

Related papers

• TopoCurate:Modeling Interaction Topology for Tool-Use Agent Training [53.93696896939915]
  Training tool-use agents typically rely on Supervised Fine-Tuning (SFT) on successful trajectories and Reinforcement Learning (RL) on pass-rate-selected tasks.<n>We propose TopoCurate, an interaction-aware framework that projects multi-trial rollouts from the same task into a unified semantic quotient topology.<n>TopoCurate achieves consistent gains of 4.2% (SFT) and 6.9% (RL) over state-of-the-art baselines.
  arXiv  Detail & Related papers  (2026-03-02T10:38:54Z)
• Embedding Poisoning: Bypassing Safety Alignment via Embedding Semantic Shift [23.0914017433021]
  This work identifies a novel class of deployment phase attacks that exploit a vulnerability by injecting imperceptible perturbations directly into the embedding layer outputs without modifying model weights or input text.<n>We propose Search based Embedding Poisoning, a practical, model agnostic framework that introduces carefully optimized perturbations into embeddings associated with high risk tokens.
  arXiv  Detail & Related papers  (2025-09-08T05:00:58Z)
• Exploiting Edge Features for Transferable Adversarial Attacks in Distributed Machine Learning [54.26807397329468]
  This work explores a previously overlooked vulnerability in distributed deep learning systems.<n>An adversary who intercepts the intermediate features transmitted between them can still pose a serious threat.<n>We propose an exploitation strategy specifically designed for distributed settings.
  arXiv  Detail & Related papers  (2025-07-09T20:09:00Z)
• Circumventing Safety Alignment in Large Language Models Through Embedding Space Toxicity Attenuation [13.971909819796762]
  Large Language Models (LLMs) have achieved remarkable success across domains such as healthcare, education, and cybersecurity.<n>Embedding space poisoning is a subtle attack vector where adversaries manipulate the internal semantic representations of input data to bypass safety alignment mechanisms.<n>We propose ETTA, a novel framework that identifies and attenuates toxicity-sensitive dimensions in embedding space via linear transformations.
  arXiv  Detail & Related papers  (2025-07-08T03:01:00Z)
• Robust Policy Switching for Antifragile Reinforcement Learning for UAV Deconfliction in Adversarial Environments [6.956559003734227]
  An unmanned aerial vehicles (UAVs) has been exposed to adversarial attacks that exploit vulnerabilities in reinforcement learning (RL)<n>This paper introduces an antifragile RL framework that enhances adaptability to broader distributional shifts.<n>It achieves superior performance, demonstrating shorter navigation path lengths and a higher rate of conflict-free navigation trajectories.
  arXiv  Detail & Related papers  (2025-06-26T10:06:29Z)
• Efficient Safety Alignment of Large Language Models via Preference Re-ranking and Representation-based Reward Modeling [84.00480999255628]
  Reinforcement Learning algorithms for safety alignment of Large Language Models (LLMs) encounter the challenge of distribution shift.<n>Current approaches typically address this issue through online sampling from the target policy.<n>We propose a new framework that leverages the model's intrinsic safety judgment capability to extract reward signals.
  arXiv  Detail & Related papers  (2025-03-13T06:40:34Z)
• Transferable Adversarial Attacks on SAM and Its Downstream Models [87.23908485521439]
  This paper explores the feasibility of adversarial attacking various downstream models fine-tuned from the segment anything model (SAM)<n>To enhance the effectiveness of the adversarial attack towards models fine-tuned on unknown datasets, we propose a universal meta-initialization (UMI) algorithm.
  arXiv  Detail & Related papers  (2024-10-26T15:04:04Z)
• One-Shot Safety Alignment for Large Language Models via Optimal Dualization [64.52223677468861]
  This paper presents a perspective of dualization that reduces constrained alignment to an equivalent unconstrained alignment problem. We do so by pre-optimizing a smooth and convex dual function that has a closed form. Our strategy leads to two practical algorithms in model-based and preference-based settings.
  arXiv  Detail & Related papers  (2024-05-29T22:12:52Z)
• SAFE-SIM: Safety-Critical Closed-Loop Traffic Simulation with Diffusion-Controllable Adversaries [94.84458417662407]
  We introduce SAFE-SIM, a controllable closed-loop safety-critical simulation framework. Our approach yields two distinct advantages: 1) generating realistic long-tail safety-critical scenarios that closely reflect real-world conditions, and 2) providing controllable adversarial behavior for more comprehensive and interactive evaluations. We validate our framework empirically using the nuScenes and nuPlan datasets across multiple planners, demonstrating improvements in both realism and controllability.
  arXiv  Detail & Related papers  (2023-12-31T04:14:43Z)
• Improving Adversarial Transferability via Intermediate-level Perturbation Decay [79.07074710460012]
  We develop a novel intermediate-level method that crafts adversarial examples within a single stage of optimization. Experimental results show that it outperforms state-of-the-arts by large margins in attacking various victim models.
  arXiv  Detail & Related papers  (2023-04-26T09:49:55Z)
• Query-Free Adversarial Transfer via Undertrained Surrogates [14.112444998191698]
  We introduce a new method for improving the efficacy of adversarial attacks in a black-box setting by undertraining the surrogate model which the attacks are generated on. We show that this method transfers well across architectures and outperforms state-of-the-art methods by a wide margin.
  arXiv  Detail & Related papers  (2020-07-01T23:12:22Z)

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.