SafeSteer: A Decoding-level Defense Mechanism for Multimodal Large Language Models

• URL: http://arxiv.org/abs/2605.11716v1
• Date: Tue, 12 May 2026 08:05:10 GMT
• Title: SafeSteer: A Decoding-level Defense Mechanism for Multimodal Large Language Models
• Abstract summary: We introduce SafeSteer, a decoding-level defense mechanism for MLLMs.<n>It includes a Decoding-Probe for detecting and correcting harmful output during decoding.<n>It can improve MLLMs' safety by up to 33.40% without fine-tuning.
• Score: 30.79900292985646
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Multimodal large language models (MLLMs) are gaining increasing attention. Due to the heterogeneity of their input features, they face significant challenges in terms of jailbreak defenses. Current defense methods rely on costly fine-tuning or inefficient post-hoc interventions, limiting their ability to address novel attacks and involving performance trade-offs. To address the above issues, we explore the inherent safety capabilities within MLLMs and quantify their intrinsic ability to discern harmfulness at decoding stage. We observe that 1) MLLMs can distinguish the harmful and harmless inputs during decoding process, 2) Image-based attacks are more stealthy. Based on these insights, we introduce SafeSteer, a decoding-level defense mechanism for MLLMs. Specifically, it includes a Decoding-Probe, a lightweight probe for detecting and correcting harmful output during decoding, which iteratively steers the decoding process toward safety. Furthermore, a modal semantic alignment vector is integrated to transfer the strong textual safety alignment to the vision modality. Experiments on multiple MLLMs demonstrate that SafeSterr can improve MLLMs' safety by up to 33.40\% without fine-tuning. Notably, it can maintain the effectiveness of MLLMs, ensuring a balance between their helpfulness and harmlessness.

Related papers

• Odysseus: Jailbreaking Commercial Multimodal LLM-integrated Systems via Dual Steganography [77.44136793431893]
  We propose a novel jailbreak paradigm that introduces dual steganography to covertly embed malicious queries into benign-looking images.<n>Our Odysseus successfully jailbreaks several pioneering and realistic MLLM-integrated systems, achieving up to 99% attack success rate.
  arXiv  Detail & Related papers  (2025-12-23T08:53:36Z)
• Q-MLLM: Vector Quantization for Robust Multimodal Large Language Model Security [12.835224376066769]
  We introduce Q-MLLM, a novel architecture that integrates two-level vector quantization to create a discrete bottleneck against adversarial attacks.<n> Experiments demonstrate that Q-MLLM achieves significantly better defense success rate against both jailbreak attacks and toxic image attacks than existing approaches.
  arXiv  Detail & Related papers  (2025-11-20T10:55:19Z)
• SafePTR: Token-Level Jailbreak Defense in Multimodal LLMs via Prune-then-Restore Mechanism [123.54980913741828]
  Multimodal Large Language Models (MLLMs) extend LLMs to support visual reasoning.<n>MLLMs are susceptible to multimodal jailbreak attacks and hindering their safe deployment.<n>We propose Safe Prune-then-Restore (SafePTR), a training-free defense framework that selectively prunes harmful tokens at vulnerable layers while restoring benign features at subsequent layers.
  arXiv  Detail & Related papers  (2025-07-02T09:22:03Z)
• Reshaping Representation Space to Balance the Safety and Over-rejection in Large Audio Language Models [50.89022445197919]
  Large Audio Language Models (LALMs) have extended the capabilities of Large Language Models (LLMs)<n>Recent research has revealed that LALMs remain vulnerable to harmful queries due to insufficient safety-alignment.
  arXiv  Detail & Related papers  (2025-05-26T08:25:25Z)
• Do We Really Need Curated Malicious Data for Safety Alignment in Multi-modal Large Language Models? [83.53005932513155]
  Multi-modal large language models (MLLMs) have made significant progress, yet their safety alignment remains limited.<n>We propose finetuning MLLMs on a small set of benign instruct-following data with responses replaced by simple, clear rejection sentences.
  arXiv  Detail & Related papers  (2025-04-14T09:03:51Z)
• CoCA: Regaining Safety-awareness of Multimodal Large Language Models with Constitutional Calibration [90.36429361299807]
  multimodal large language models (MLLMs) have demonstrated remarkable success in engaging in conversations involving visual inputs. The integration of visual modality has introduced a unique vulnerability: the MLLM becomes susceptible to malicious visual inputs. We introduce a technique termed CoCA, which amplifies the safety-awareness of the MLLM by calibrating its output distribution.
  arXiv  Detail & Related papers  (2024-09-17T17:14:41Z)
• Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformation [98.02846901473697]
  We propose ECSO (Eyes Closed, Safety On), a training-free protecting approach that exploits the inherent safety awareness of MLLMs. ECSO generates safer responses via adaptively transforming unsafe images into texts to activate the intrinsic safety mechanism of pre-aligned LLMs.
  arXiv  Detail & Related papers  (2024-03-14T17:03:04Z)
• MLLM-Protector: Ensuring MLLM's Safety without Hurting Performance [36.03512474289962]
  This paper investigates the novel challenge of defending MLLMs against malicious attacks through visual inputs. Images act as a foreign language" that is not considered during safety alignment, making MLLMs more prone to producing harmful responses. We introduce MLLM-Protector, a plug-and-play strategy that solves two subtasks: 1) identifying harmful responses via a lightweight harm detector, and 2) transforming harmful responses into harmless ones via a detoxifier.
  arXiv  Detail & Related papers  (2024-01-05T17:05:42Z)

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.