YuFeng-XGuard: A Reasoning-Centric, Interpretable, and Flexible Guardrail Model for Large Language Models

TL;DR

YuFeng-XGuard redefines LLM safety guardrails as a reasoning-centric, multi-dimensional risk perception system that outputs explicit risk categories, calibrated confidences, and natural language explanations. It employs a tiered inference approach to provide instant first-token decisions with on-demand deeper reasoning, and a dynamic policy framework that decouples perception from enforcement, enabling agile policy updates without retraining. Trained with a large, multilingual, reasoning-focused corpus and a structured SFT objective, it achieves state-of-the-art results across generic safety, multilingual robustness, jailbreak resistance, and safe-completion benchmarks, while offering a lightweight 0.6B variant for latency-constrained deployments. The work demonstrates strong practical impact by delivering interpretable, configurable, and scalable guardrails suitable for real-world, diverse-language AI applications, with open releases to foster further research and deployment variation.

Abstract

As large language models (LLMs) are increasingly deployed in real-world applications, safety guardrails are required to go beyond coarse-grained filtering and support fine-grained, interpretable, and adaptable risk assessment. However, existing solutions often rely on rapid classification schemes or post-hoc rules, resulting in limited transparency, inflexible policies, or prohibitive inference costs. To this end, we present YuFeng-XGuard, a reasoning-centric guardrail model family designed to perform multi-dimensional risk perception for LLM interactions. Instead of producing opaque binary judgments, YuFeng-XGuard generates structured risk predictions, including explicit risk categories and configurable confidence scores, accompanied by natural language explanations that expose the underlying reasoning process. This formulation enables safety decisions that are both actionable and interpretable. To balance decision latency and explanatory depth, we adopt a tiered inference paradigm that performs an initial risk decision based on the first decoded token, while preserving ondemand explanatory reasoning when required. In addition, we introduce a dynamic policy mechanism that decouples risk perception from policy enforcement, allowing safety policies to be adjusted without model retraining. Extensive experiments on a diverse set of public safety benchmarks demonstrate that YuFeng-XGuard achieves stateof-the-art performance while maintaining strong efficiency-efficacy trade-offs. We release YuFeng-XGuard as an open model family, including both a full-capacity variant and a lightweight version, to support a wide range of deployment scenarios.

Figures (2)

• Figure 1: Top average F1 scores across classification benchmarks
• Figure 2: Overall win rate of across all benchmarks.