DriftGuard: Safety-Aware Multi-Monitor Detection and Selective Adaptation for Evolving Toxicity Moderation

2026-06-27

AI Summary

This work addresses the challenge of online toxic content moderation in the face of evolving harmful behaviors—such as the use of coded language and shifting attack targets—which often evade detection by conventional global drift monitoring approaches that fail to capture safety-critical changes within localized high-risk subspaces. To this end, the authors propose DriftGuard, a novel framework that integrates safety-sensitive signals including identity-based harm, false negative risk, and model uncertainty into a multi-monitor architecture for tracking multidimensional distributional shifts. DriftGuard further introduces a hard-mixture selective fine-tuning strategy that prioritizes high-risk samples for lightweight model adaptation. Evaluated on the Civil Comments and DynaHate datasets, the method achieves substantial improvements in toxicity recall (0.8777 and 0.8523, respectively) and reduces the false negative rate by 0.0781, demonstrating its robustness and effectiveness in dynamic moderation environments.

Abstract

Automated toxicity moderation systems operate in dynamic online environments where harmful behavior evolves through coded language, shifting targets, and strategic adaptation to enforcement. Existing drift detection methods often focus on global distributional change, but such signals may miss safety-relevant shifts that emerge in localized harm subspaces or high-risk model-error regions. This paper introduces DriftGuard, a safety-aware adaptive moderation framework that combines multi-monitor drift detection with selective model updating. The framework tracks global text drift, identity-harm drift, model uncertainty, toxic-risk drift, and false-negative-risk drift. When safety-relevant change is detected, the model is updated using a hard-mix adaptation set that prioritizes likely false negatives, identity-related high-risk examples, false-positive-risk examples, and uncertain boundary cases. Experiments on Civil Comments temporal shift and Jigsaw-to-DynaHate cross-dataset shift show that safety-aware monitors detect risks missed by global drift alone. Hard-mix adaptation improves toxic recall and accuracy over no-update and random-balanced baselines, raising toxic recall to 0.8777 on Civil Comments and from 0.7107 to 0.8523 on DynaHate. Bootstrap analysis further shows stable DynaHate safety gains, with toxic recall increasing by 0.1418 and false-negative prevalence decreasing by 0.0781. Overall, DriftGuard links safety-aware drift detection to targeted, lightweight model updating for more robust adaptive toxicity moderation.

Problem

• toxicity moderation
• distributional drift
• safety-aware detection
• model adaptation
• false negatives

Innovation

• safety-aware drift detection
• multi-monitor framework
• selective adaptation
• hard-mix adaptation
• toxicity moderation