ThinkGuard：审慎的慢思考带来谨慎的护栏
ThinkGuard: Deliberative Slow Thinking Leads to Cautious Guardrails
AI总结：
本文提出 ThinkGuard，一种批评增强的护栏模型，通过从高能力 LLMs 蒸馏结构化批评与安全标签，显著提升护栏的谨慎性与可解释性，在多个安全基准上取得最优 F1 与 AUPRC，准确率较 LLaMA Guard 3 提升 16.1%。
摘要：
随着大语言模型（LLMs）部署到实际应用中，确保其安全性至关重要。现有的护栏依赖基于规则的过滤或单次分类，限制了其处理细微安全违规的能力。为此，我们提出 ThinkGuard，这是一种批评增强的护栏模型，通过在生成安全标签的同时生成结构化批评，从高能力 LLMs 中蒸馏知识。在批评增强数据上微调后，所捕获的审慎思考能力极大地提升了护栏的谨慎性和可解释性。在多个安全基准上评估，ThinkGuard 取得了最高的平均 F1 和 AUPRC，优于所有基线。与 LLaMA Guard 3 相比，ThinkGuard 将准确率提高了 16.1%，宏 F1 提高了 27.0%。此外，它超越了仅标签微调的模型，证实结构化批评在保持计算效率的同时提升了分类精度和细微安全推理能力。
英文摘要：
Ensuring the safety of large language models (LLMs) is critical as they are deployed in real-world applications. Existing guardrails rely on rule-based filtering or single-pass classification, limiting their ability to handle nuanced safety violations. To address this, we propose ThinkGuard, a critique-augmented guardrail model that distills knowledge from high-capacity LLMs by generating structured critiques alongside safety labels. Fine-tuned on critique-augmented data, the captured deliberative thinking ability drastically enhances the guardrail's cautiousness and interpretability. Evaluated on multiple safety benchmarks, ThinkGuard achieves the highest average F1 and AUPRC, outperforming all baselines. Compared to LLaMA Guard 3, ThinkGuard improves accuracy by 16.1% and macro F1 by 27.0%. Moreover, it surpasses label-only fine-tuned models, confirming that structured critiques enhance both classification precision and nuanced safety reasoning while maintaining computational efficiency.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
