SGM: Safety Glasses for Multimodal Large Language Models via Neuron-Level Detoxification

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Hongbo, AprilPyone, MaungMaung, Echizen, Isao
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908830960451584
author Wang, Hongbo
AprilPyone, MaungMaung
Echizen, Isao
author_facet Wang, Hongbo
AprilPyone, MaungMaung
Echizen, Isao
contents Disclaimer: Samples in this paper may be harmful and cause discomfort. Multimodal large language models (MLLMs) enable multimodal generation but inherit toxic, biased, and NSFW signals from weakly curated pretraining corpora, causing safety risks, especially under adversarial triggers that late, opaque training-free detoxification methods struggle to handle. We propose SGM, a white-box neuron-level multimodal intervention that acts like safety glasses for toxic neurons: it selectively recalibrates a small set of toxic expert neurons via expertise-weighted soft suppression, neutralizing harmful cross-modal activations without any parameter updates. We establish MM-TOXIC-QA, a multimodal toxicity evaluation framework, and compare SGM with existing detoxification techniques. Experiments on open-source MLLMs show that SGM mitigates toxicity in standard and adversarial conditions, cutting harmful rates from 48.2\% to 2.5\% while preserving fluency and multimodal reasoning. SGM is extensible, and its combined defenses, denoted as SGM*, integrate with existing detoxification methods for stronger safety performance, providing an interpretable, low-cost solution for toxicity-controlled multimodal generation.
format Preprint
id arxiv_https___arxiv_org_abs_2512_15052
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SGM: Safety Glasses for Multimodal Large Language Models via Neuron-Level Detoxification
Wang, Hongbo
AprilPyone, MaungMaung
Echizen, Isao
Computation and Language
Artificial Intelligence
Disclaimer: Samples in this paper may be harmful and cause discomfort. Multimodal large language models (MLLMs) enable multimodal generation but inherit toxic, biased, and NSFW signals from weakly curated pretraining corpora, causing safety risks, especially under adversarial triggers that late, opaque training-free detoxification methods struggle to handle. We propose SGM, a white-box neuron-level multimodal intervention that acts like safety glasses for toxic neurons: it selectively recalibrates a small set of toxic expert neurons via expertise-weighted soft suppression, neutralizing harmful cross-modal activations without any parameter updates. We establish MM-TOXIC-QA, a multimodal toxicity evaluation framework, and compare SGM with existing detoxification techniques. Experiments on open-source MLLMs show that SGM mitigates toxicity in standard and adversarial conditions, cutting harmful rates from 48.2\% to 2.5\% while preserving fluency and multimodal reasoning. SGM is extensible, and its combined defenses, denoted as SGM*, integrate with existing detoxification methods for stronger safety performance, providing an interpretable, low-cost solution for toxicity-controlled multimodal generation.
title SGM: Safety Glasses for Multimodal Large Language Models via Neuron-Level Detoxification
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2512.15052