Taxonomy-Adaptive Moderation Model with Robust Guardrails for Large Language Models

TL;DR

Roblox Guard 1.0 is a fine-tuned LLM designed to improve safety by robustly moderating inputs and outputs, generalizing across unseen safety taxonomies, and evaluated with a new extensible safety benchmark.

Contribution

The paper introduces Roblox Guard 1.0, a novel instruction fine-tuned LLM with comprehensive moderation capabilities and a new safety evaluation benchmark, enhancing safety across diverse scenarios.

Findings

Roblox Guard 1.0 outperforms existing safety benchmarks.

The model generalizes well to unseen safety taxonomies.

The new benchmark effectively evaluates moderation robustness.

Abstract

Large Language Models (LLMs) are typically aligned for safety during the post-training phase; however, they may still generate inappropriate outputs that could potentially pose risks to users. This challenge underscores the need for robust safeguards that operate across both model inputs and outputs. In this work, we introduce Roblox Guard 1.0, a state-of-the-art instruction fine-tuned LLM designed to enhance the safety of LLM systems through comprehensive input-output moderation, using a pipeline of LLMs to enhance moderation capability. Built on the Llama-3.1-8B-Instruct backbone, our model is instruction fine-tuned to generalize across previously unseen safety taxonomies and demonstrates strong performance on out-of-domain safety benchmarks. The instruction fine-tuning process uses a mix of synthetic and open-source safety datasets, augmented with chain-of-thought (CoT) rationales and input inversion to enhance contextual understanding and decision making. To support systematic evaluation, we also release RobloxGuard-Eval, a new benchmark featuring an extensible safety taxonomy to assess the effectiveness of LLM guardrails and moderation frameworks.