Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Mitigating the Safety Alignment Tax with Null-Space Constrained Policy Optimization

About

As Large Language Models (LLMs) are increasingly deployed in real-world applications, it is important to ensure their behaviors align with human values, societal norms, and ethical principles. However, safety alignment under Reinforcement Learning (RL) often suffers from forgetting learned general abilities, which is also known as the alignment tax. To address this issue, we introduce Null-Space constrained Policy Optimization (NSPO), a novel RL framework for LLM safety alignment while preserving their core abilities. The safety policy gradients are geometrically projected into the null space of general tasks, thereby mitigating the safety alignment tax. In addition, we theoretically prove that NSPO preserves the model's original core capabilities, while still guaranteeing a descent direction for effective safety alignment. Extensive experiments demonstrate that NSPO outperforms existing methods by a large margin, achieving state-of-the-art safety performance without sacrificing accuracy on general tasks, including math, code, and instruction-following tasks. Notably, NSPO is data-efficient and only requires 40% of public human-annotated safety data from PKU-SafeRLHF to achieve promising safety performance, without a large amount of mixed general tasks data in existing alignment methods.

Yifan Niu, Han Xiao, Dongyi Liu, Nuo Chen, Jia Li• 2025

Related benchmarks

TaskDatasetResultRank
Instruction FollowingAlpacaEval
Win Rate94.33
423
General KnowledgeMMLU
MMLU General Knowledge Accuracy71.74
373
Jailbreak RobustnessWildJailbreak
Unsafe Rate4.89
165
Math ReasoningMATH
Accuracy75.05
160
Math ReasoningMATH500
Accuracy57.4
127
Over-refusalXSTest--
102
Scientific ReasoningGPQA Diamond
Accuracy36.87
94
Code GenerationLiveCodeBench
Pass@10.2405
86
Math ReasoningOlympiadBench
Accuracy39
76
Harmful Request DefenseAdvBench
ASR0.06
65
Showing 10 of 22 rows

Other info

Follow for update