SafeNexus: Discovering and Steering Modality-Universal Safety Neurons in MLLMs

2026-07-30

AI Summary

This work addresses the limitations of existing safety mechanisms in multimodal large language models (MLLMs), which are predominantly confined to single modalities and thus ill-equipped to handle cross-modal security threats. The authors propose SafeNexus, a novel framework that identifies and validates universal safety neurons (US-Neurons)—neural units shared across modalities that govern safe behavior. Leveraging activation amplification and targeted calibration as neuron-level intervention strategies, SafeNexus achieves cross-modal safety alignment through intermediate-layer activation analysis, importance scoring, and contrastive data-driven neuron localization, complemented by precise suppression and fine-tuning. Experimental results demonstrate that SafeNexus significantly outperforms state-of-the-art baselines across diverse multimodal safety benchmarks while preserving overall model performance.

Abstract

Although Large Language Models (LLMs) have demonstrated promising safety performance, extending them to Multimodal Large Language Models (MLLMs) exposes a significant gap between expanded multimodal capabilities and existing safety mechanisms. Current defenses remain predominantly confined to specific modal settings, thereby limiting their robustness against broader cross-modal threats. To bridge this gap, we introduce SafeNexus, a cross-modal safety alignment framework that adopts a dedicated neuron-level intervention strategy. First, we formulate a neuron localization paradigm that identifies functionally specialized neurons by characterizing intermediate-layer activation patterns and quantifying their functional salience through importance scoring. Building upon this paradigm, we exploit contrastive data to identify modality-bound safety neurons (BS-Neurons), and validate their role in regulating safety behavior within each modality via targeted suppression. Further cross-modal analysis defines modality-universal safety neurons (US-Neurons) as the shared subset of BS-Neurons identified across individual modalities, serving as the core for defending against harmful cross-modal attacks. We observe that suppressing these neurons substantially degrades safety performance across modalities, while leaving overall utility largely unaffected. Building on these insights, we propose two safety alignment strategies: activation-level safety amplifier and safety neuron calibrator. The proposed strategies enhance model safety through two distinct routes: the former amplifies the activation magnitudes of US-Neurons, while the latter selectively calibrates them via targeted fine-tuning. Extensive experiments demonstrate that our method outperforms prevailing state-of-the-art approaches on safety benchmarks spanning diverse modality combinations, while effectively preserving utility.

Problem

• Multimodal Large Language Models
• safety mechanisms
• cross-modal threats
• modality-universal safety

Innovation

• modality-universal safety neurons
• neuron-level intervention
• cross-modal safety alignment
• safety neuron calibration
• multimodal large language models