ThinkGuard: Deliberative Slow Thinking Leads to Cautious Guardrails
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910970912178176 |
|---|---|
| author | Wen, Xiaofei Zhou, Wenxuan Mo, Wenjie Jacky Chen, Muhao |
| author_facet | Wen, Xiaofei Zhou, Wenxuan Mo, Wenjie Jacky Chen, Muhao |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_13458 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | ThinkGuard: Deliberative Slow Thinking Leads to Cautious Guardrails Wen, Xiaofei Zhou, Wenxuan Mo, Wenjie Jacky Chen, Muhao Computation and Language Artificial Intelligence Cryptography and Security Machine Learning 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. |
| title | ThinkGuard: Deliberative Slow Thinking Leads to Cautious Guardrails |
| topic | Computation and Language Artificial Intelligence Cryptography and Security Machine Learning |
| url | https://arxiv.org/abs/2502.13458 |