FlexGuard: Continuous Risk Scoring for Strictness-Adaptive LLM Content Moderation

TL;DR

FlexGuard introduces a continuous risk scoring system for LLM content moderation, enabling adaptable strictness levels and improving robustness across different enforcement regimes.

Contribution

The paper presents FlexBench, a new benchmark for strictness-adaptive moderation, and FlexGuard, a model that outputs calibrated risk scores for flexible, robust content moderation.

Findings

FlexGuard outperforms existing models under varying strictness regimes.

Models trained with risk-alignment show improved score-severity consistency.

FlexGuard achieves higher accuracy and robustness on FlexBench and public benchmarks.

Abstract

Ensuring the safety of LLM-generated content is essential for real-world deployment. Most existing guardrail models formulate moderation as a fixed binary classification task, implicitly assuming a fixed definition of harmfulness. In practice, enforcement strictness - how conservatively harmfulness is defined and enforced - varies across platforms and evolves over time, making binary moderators brittle under shifting requirements. We first introduce FlexBench, a strictness-adaptive LLM moderation benchmark that enables controlled evaluation under multiple strictness regimes. Experiments on FlexBench reveal substantial cross-strictness inconsistency in existing moderators: models that perform well under one regime can degrade substantially under others, limiting their practical usability. To address this, we propose FlexGuard, an LLM-based moderator that outputs a calibrated continuous risk score reflecting risk severity and supports strictness-specific decisions via thresholding. We train FlexGuard via risk-alignment optimization to improve score-severity consistency and provide practical threshold selection strategies to adapt to target strictness at deployment. Experiments on FlexBench and public benchmarks demonstrate that FlexGuard achieves higher moderation accuracy and substantially improved robustness under varying strictness. We release the source code and data to support reproducibility.