STAR-1: Safer Alignment of Reasoning LLMs with 1K Data
Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (44) · 37988-37997
DOI: 10.1609/aaai.v40i44.41136 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Zijun Wang und 8 weiteren Autor:innen. Die Arbeit wurde bisher 4-mal zitiert (Crossref).
Abstract (Originalsprache)
This paper introduces STAR-1, a high-quality, just-1k-scale safety dataset specifically designed for large reasoning models (LRMs) like DeepSeek-R1. Built on three core principles --- diversity, deliberative reasoning, and rigorous filtering --- STAR-1 aims to address the critical needs for safety alignment in LRMs. Specifically, we begin by integrating existing open-source safety datasets from diverse sources. Then, we curate safety policies to generate policy-grounded deliberative reasoning samples. Lastly, we apply a GPT-4o-based safety scoring system to select training examples aligned with best practices. Experimental results show that fine-tuning LRMs with STAR-1 leads to an average 40% improvement in safety performance across four benchmarks, while only incurring a marginal decrease (e.g., an average of 1.1%) in reasoning ability measured across five reasoning tasks. Extensive ablation studies further validate the importance of our design principles in constructing STAR-1 and analyze its efficacy across both LRMs and traditional LLMs.
Zitieren
APA
Wang, Z., Tu, H., Wang, Y., Wu, J., Liu, Y., Mei, J., Bartoldson, B. R., Kailkhura, B., & Xie, C. (2026). STAR-1: Safer Alignment of Reasoning LLMs with 1K Data. Proceedings of the AAAI Conference on Artificial Intelligence, 40(44), 37988-37997. https://doi.org/10.1609/aaai.v40i44.41136
Vancouver
Wang Z, Tu H, Wang Y, Wu J, Liu Y, Mei J, et al. STAR-1: Safer Alignment of Reasoning LLMs with 1K Data. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(44):37988-37997. doi:10.1609/aaai.v40i44.41136
BibTeX
@article{Wang2026,
author = {Wang, Zijun and Tu, Haoqin and Wang, Yuhan and Wu, Juncheng and Liu, Yanqing and Mei, Jieru and Bartoldson, Brian R. and Kailkhura, Bhavya and Xie, Cihang},
title = {STAR-1: Safer Alignment of Reasoning LLMs with 1K Data},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2026},
volume = {40},
number = {44},
pages = {37988--37997},
doi = {10.1609/aaai.v40i44.41136}
}
EUVIMED