Diagnosing and Repairing Unsafe Channels in Vision-Language Models via Causal Discovery and Dual-Modal Safety Subspace Projection

• URL: http://arxiv.org/abs/2603.27240v1
• Date: Sat, 28 Mar 2026 11:31:16 GMT
• Title: Diagnosing and Repairing Unsafe Channels in Vision-Language Models via Causal Discovery and Dual-Modal Safety Subspace Projection
• Abstract summary: We present a comprehensive framework for diagnosing and repairing unsafe channels within Large Vision-Language Models (LVLMs)<n>We first perform causal mediation analysis to identify neurons and layers that are causally responsible for unsafe behaviors.<n>We introduce a dual-modal safety subspace projection method that learns generalized safety subspaces for both visual and textual modalities.
• Score: 21.411964269762432
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large Vision-Language Models (LVLMs) have achieved impressive performance across multimodal understanding and reasoning tasks, yet their internal safety mechanisms remain opaque and poorly controlled. In this work, we present a comprehensive framework for diagnosing and repairing unsafe channels within LVLMs (CARE). We first perform causal mediation analysis to identify neurons and layers that are causally responsible for unsafe behaviors. Based on these findings, we introduce a dual-modal safety subspace projection method that learns generalized safety subspaces for both visual and textual modalities through generalized eigen-decomposition between benign and malicious activations. During inference, activations are dynamically projected toward these safety subspaces via a hybrid fusion mechanism that adaptively balances visual and textual corrections, effectively suppressing unsafe features while preserving semantic fidelity. Extensive experiments on multiple safety benchmarks demonstrate that our causal-subspace repair framework significantly enhances safety robustness without degrading general multimodal capabilities, outperforming prior activation steering and alignment-based baselines. Additionally, our method exhibits good transferability, defending against unseen attacks.

Related papers

• SafeNexus: Discovering and Steering Modality-Universal Safety Neurons in MLLMs [69.51621286001262]
  We introduce SafeNexus, a cross-modal safety alignment framework that adopts a dedicated neuron-level intervention strategy.<n>We identify modality-bound safety neurons (BS-Neurons), and validate their role in regulating safety behavior within each modality.<n>We propose two safety alignment strategies: activation-level safety amplifier and safety neuron calibrator.
  arXiv  Detail & Related papers  (2026-07-31T02:45:56Z)
• RefusalGuard: Geometry-Preserving Fine-Tuning for Safety in LLMs [5.100622189286672]
  Fine-tuning safety-aligned language models for downstream tasks can lead to substantial degradation of refusal behavior.<n>We introduce REFUSALGUARD, a representation-level fine-tuning framework that preserves safety-relevant structure during model adaptation.
  arXiv  Detail & Related papers  (2026-05-03T14:48:18Z)
• Unsafe by Reciprocity: How Generation-Understanding Coupling Undermines Safety in Unified Multimodal Models [53.41258113970795]
  We investigate whether cross-functionality reciprocity itself constitutes a structural source of vulnerability in Unified Multimodal Models (UMMs)<n>We propose RICE: Reciprocal Interaction-based Cross-functionality Exploitation, a novel attack paradigm that explicitly exploits bidirectional interactions between understanding and generation.
  arXiv  Detail & Related papers  (2026-03-28T16:28:39Z)
• Enhancing Safety of Large Language Models via Embedding Space Separation [24.821172423424866]
  Large language models (LLMs) have achieved impressive capabilities, yet ensuring their safety against harmful prompts remains a critical challenge.<n>Recent work has revealed that the latent representations (embeddings) of harmful and safe queries in LLMs typically exhibit linear separability.<n>Motivated by this observation, we propose a representation-level fine-tuning approach, named Embedding Space Separation (ES2)
  arXiv  Detail & Related papers  (2026-03-01T07:22:38Z)
• BarrierSteer: LLM Safety via Learning Barrier Steering [83.12893815611052]
  BarrierSteer is a novel framework that formalizes safety by embedding learned non-linear safety constraints directly into the model's latent representation space.<n>We show that BarrierSteer substantially reduces adversarial success rates, decreases unsafe generations, and outperforms existing methods.
  arXiv  Detail & Related papers  (2026-02-23T18:19:46Z)
• Self-Guard: Defending Large Reasoning Models via enhanced self-reflection [54.775612141528164]
  Self-Guard is a lightweight safety defense framework for Large Reasoning Models.<n>It bridges the awareness-compliance gap, achieving robust safety performance without compromising model utility.<n>Self-Guard exhibits strong generalization across diverse unseen risks and varying model scales.
  arXiv  Detail & Related papers  (2026-01-31T13:06:11Z)
• Attributing and Exploiting Safety Vectors through Global Optimization in Large Language Models [50.91504059485288]
  We propose a framework that identifies safety-critical attention heads through global optimization over all heads simultaneously.<n>We develop a novel inference-time white-box jailbreak method that exploits the identified safety vectors through activation repatching.
  arXiv  Detail & Related papers  (2026-01-22T09:32:43Z)
• Breaking the Safety-Capability Tradeoff: Reinforcement Learning with Verifiable Rewards Maintains Safety Guardrails in LLMs [3.198812241868092]
  reinforcement learning with verifiable rewards (RLVR) has emerged as a promising alternative that optimize models on objectively measurable tasks.<n>We present the first comprehensive theoretical and empirical analysis of safety properties in RLVR.<n> Empirically, we conduct extensive experiments across five adversarial safety benchmarks, demonstrating that RLVR can simultaneously enhance reasoning capabilities while maintaining or improving safety guardrails.
  arXiv  Detail & Related papers  (2025-11-26T04:36:34Z)
• LatentGuard: Controllable Latent Steering for Robust Refusal of Attacks and Reliable Response Generation [4.29885665563186]
  LATENTGUARD is a framework that combines behavioral alignment with supervised latent space control for interpretable and precise safety steering.<n>Our results show significant improvements in both safety controllability and response interpretability without compromising utility.
  arXiv  Detail & Related papers  (2025-09-24T07:31:54Z)
• Self-Aware Safety Augmentation: Leveraging Internal Semantic Understanding to Enhance Safety in Vision-Language Models [21.961325147038867]
  Large vision-language models (LVLMs) are vulnerable to harmful input compared to their language-only backbones.<n>We define these capabilities as safety perception, semantic understanding, and alignment for linguistic expression.<n>Motivated by these findings, we propose textbfSelf-Aware Safety Augmentation (SASA), a technique that projects informative semantic representations onto earlier safety-oriented layers.
  arXiv  Detail & Related papers  (2025-07-29T09:48:57Z)
• HoliSafe: Holistic Safety Benchmarking and Modeling for Vision-Language Model [58.12612140992874]
  We introduce a holistic safety dataset and benchmark, textbfHoliSafe, that spans all five safe/unsafe image-text combinations.<n>We also propose a novel modular framework for enhancing VLM safety with a visual guard module (VGM) designed to assess the harmfulness of input images.<n> Experiments show that Safe-VLM with VGM, trained on our HoliSafe, achieves state-of-the-art safety performance across multiple VLM benchmarks.
  arXiv  Detail & Related papers  (2025-06-05T07:26:34Z)
• Understanding and Rectifying Safety Perception Distortion in VLMs [19.239094089025095]
  Vision-language models (VLMs) become more susceptible to harmful requests and jailbreak attacks after integrating the vision modality.<n> multimodal inputs introduce an modality-induced activation shift toward a "safer" direction compared to their text-only counterparts.<n>We propose ShiftDC, a training-free method that decomposes and calibrates the modality-induced activation shift to reduce the impact of modality on safety.
  arXiv  Detail & Related papers  (2025-02-18T18:06:48Z)

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.