AlphaAlign: Incentivizing Safety Alignment with Extremely Simplified Reinforcement Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Yi, Zhang, An, Zhang, XiuYu, Sheng, Leheng, Chen, Yuxin, Liang, Zhenkai, Wang, Xiang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913950153572352
author Zhang, Yi
Zhang, An
Zhang, XiuYu
Sheng, Leheng
Chen, Yuxin
Liang, Zhenkai
Wang, Xiang
author_facet Zhang, Yi
Zhang, An
Zhang, XiuYu
Sheng, Leheng
Chen, Yuxin
Liang, Zhenkai
Wang, Xiang
contents Large language models (LLMs), despite possessing latent safety understanding from their vast pretraining data, remain vulnerable to generating harmful content and exhibit issues such as over-refusal and utility degradation after safety alignment. Current safety alignment methods often result in superficial refusal shortcuts or rely on intensive supervision for reasoning-based approaches, failing to fully leverage the model's intrinsic safety self-awareness. We propose \textbf{AlphaAlign}, a simple yet effective pure reinforcement learning (RL) framework with verifiable safety reward designed to incentivize this latent safety awareness through proactive safety reasoning.} AlphaAlign employs a dual-reward system: a verifiable safety reward encourages correctly formatted and explicitly justified refusals for harmful queries while penalizing over-refusals, and a normalized helpfulness reward guides high-quality responses to benign inputs. This allows the model to develop proactive safety reasoning capabilities without depending on supervised safety-specific reasoning data. AlphaAlign demonstrates three key advantages: (1) Simplicity and efficiency, requiring only binary prompt safety labels and minimal RL steps for substantial improvements. (2) Breaking the safety-utility trade-off, by enhancing refusal of harmful content and reducing over-refusals, while simultaneously maintaining or even improving general task performance and robustness to unseen jailbreaks. (3) Deep alignment, fostering proactive safety reasoning that generates explicit safety rationales rather than relying on shallow refusal patterns.
format Preprint
id arxiv_https___arxiv_org_abs_2507_14987
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle AlphaAlign: Incentivizing Safety Alignment with Extremely Simplified Reinforcement Learning
Zhang, Yi
Zhang, An
Zhang, XiuYu
Sheng, Leheng
Chen, Yuxin
Liang, Zhenkai
Wang, Xiang
Artificial Intelligence
Cryptography and Security
Machine Learning
Large language models (LLMs), despite possessing latent safety understanding from their vast pretraining data, remain vulnerable to generating harmful content and exhibit issues such as over-refusal and utility degradation after safety alignment. Current safety alignment methods often result in superficial refusal shortcuts or rely on intensive supervision for reasoning-based approaches, failing to fully leverage the model's intrinsic safety self-awareness. We propose \textbf{AlphaAlign}, a simple yet effective pure reinforcement learning (RL) framework with verifiable safety reward designed to incentivize this latent safety awareness through proactive safety reasoning.} AlphaAlign employs a dual-reward system: a verifiable safety reward encourages correctly formatted and explicitly justified refusals for harmful queries while penalizing over-refusals, and a normalized helpfulness reward guides high-quality responses to benign inputs. This allows the model to develop proactive safety reasoning capabilities without depending on supervised safety-specific reasoning data. AlphaAlign demonstrates three key advantages: (1) Simplicity and efficiency, requiring only binary prompt safety labels and minimal RL steps for substantial improvements. (2) Breaking the safety-utility trade-off, by enhancing refusal of harmful content and reducing over-refusals, while simultaneously maintaining or even improving general task performance and robustness to unseen jailbreaks. (3) Deep alignment, fostering proactive safety reasoning that generates explicit safety rationales rather than relying on shallow refusal patterns.
title AlphaAlign: Incentivizing Safety Alignment with Extremely Simplified Reinforcement Learning
topic Artificial Intelligence
Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2507.14987