cs.CLNov 9, 2026

CaseLaw: A Policy-Decoupled Framework for Training-Free Content Moderation

Authors: Qi Lu, Sen Li, Likun Shi, Xiaolong Jiang

Organizations: Xiaohongshu · University of Science and Technology of China

Abstract

Content moderation is inherently policy-dependent: moderation boundaries are often subtle, subjective, and frequently updated, while conventional rule based or fine-tuning based systems require costly rule rewriting, model retraining, or system redeployment. We propose CaseLaw, a policy-decoupled and training-free framework for adaptive content moderation. CaseLaw reformulates moderation as precedent-based retrieval and voting, where multimodal content is transformed into evidence-preserved structured semantics, embedded into a stable semantic space, and classified by thresholded voting over retrieved precedent labels. We train NoteCap on a large-scale corpus of structurally annotated notes to ensure consistent semantic extraction and freeze it during subsequent policy adaptation, which requires no additional training. Policy adaptation is achieved by updating the semantic case library, including case labels and mappings, without retraining models or modifying prompts. To evaluate CaseLaw, we construct MFO-RedNotes, a benchmark for the male-female opposition risk domain. We additionally provide a five category CoPro comparison. Experiments show that CaseLaw consistently outperforms competing baselines in both precision and recall across policy-dependent risk levels and broader safety categories, with category-wise precision/recall reaching up to 86.8%/90.1% on MFO-RedNotes across both policy settings and 94.5%/92.3% on CoPro. Our work highlights the potential of precedent-based reasoning for policy adaptive content moderation and offers a practical direction toward more stable, transparent, and training-free moderation systems.

Figures & tables

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Sep 9, 2026

Can Foundation Models Moderate Online Content? Evaluating Instruction- vs. Example-Driven Policy Operationalization

The growing complexity of content moderation policies presents a critical challenge for their consistent operationalization. While foundation models possess the basic capabilities needed to confront this challenge, whether they can reliably moderate online content remains an unanswered question. In this paper, we systematically compare two competing paradigms for Vision-Language Model (VLM) guidance: an instruction-driven approach where models reason from policy precepts, and an example-driven approach where they generalize from prior precedents. We ground this investigation in ModerationBench, a new benchmark of 4,000 manually annotated, in-the-wild posts from the Bluesky platform. Our experiments reveal that foundation models can substantially outperform Bluesky's deployed moderation system, nearly tripling its F1F_1 score (0.60 vs. 0.22) on Random Posts in the benchmark, with both instruction- and example-driven paradigms achieving comparable peak effectiveness. Our findings thus chart a path toward reliable and adaptable policy operationalization at scale.
May 8, 2026

RuleSafe-VL: Evaluating Rule-Conditioned Decision Reasoning in Vision-Language Content Moderation

Platform content moderation applies explicit policy rules and context-dependent conditions to decide whether user content is allowed, restricted, or removed. A correct moderation outcome must therefore depend on which rules a case activates, how those rules interact, and whether the available evidence is sufficient. Current multimodal safety benchmarks largely reduce moderation to matching predefined final labels, leaving this underlying rule structure untested. As a result, a high benchmark score reveals little about whether a model applies the policy correctly or arrives at the correct label through superficial cues. To evaluate this rule-governed process, we introduce RuleSafe-VL, a benchmark for rule-conditioned decision reasoning in vision-language content moderation. Derived from publicly available platform moderation policies, RuleSafe-VL formalizes 93 atomic rules and 92 typed rule relations, yielding 2,166 context-sensitive image-text cases across three high-risk policy families. Its four diagnostic tasks decompose moderation into a rule-conditioned decision chain. They identify activated rules, recover rule interactions, judge decision sufficiency, and resolve outcomes once missing context is supplied. Experiments on 10 frontier, open-source, and safety-oriented VLMs reveal rule-relation recovery as the dominant bottleneck, where the best model reaches only 64.8 Macro-F1 and some safety-oriented models fall below 7 Macro-F1. Decision-state prediction also remains unreliable, peaking at 64.5 Macro-F1. RuleSafe-VL shifts moderation evaluation from final-label scoring toward diagnostic assessment of rule-conditioned decision reasoning.
Mar 15, 2026

Tool-MCoT: Tool Augmented Multimodal Chain-of-Thought for Content Safety Moderation

The growth of online platforms and user content requires strong content moderation systems that can handle complex inputs from various media types. While large language models (LLMs) are effective, their high computational cost and latency present significant challenges for scalable deployment. To address this, we introduce Tool-MCoT, a small language model (SLM) fine-tuned for content safety moderation leveraging external framework. By training our model on tool-augmented chain-of-thought data generated by LLM, we demonstrate that the SLM can learn to effectively utilize these tools to improve its reasoning and decision-making. Our experiments show that the fine-tuned SLM achieves significant performance gains. Furthermore, we show that the model can learn to use these tools selectively, achieving a balance between moderation accuracy and inference efficiency by calling tools only when necessary.