AlphaAlign: Incentivizing Safety Alignment with Extremely Simplified Reinforcement Learning

TL;DR

AlphaAlign introduces a simple reinforcement learning framework that enhances safety reasoning in large language models, effectively balancing harmful content refusal and utility preservation without extensive supervision.

Contribution

It presents a minimalistic RL approach with verifiable safety rewards that promotes proactive safety reasoning, surpassing traditional methods in efficiency and effectiveness.

Findings

Improves safety refusal accuracy with minimal RL steps.

Reduces over-refusals while maintaining task performance.

Fosters explicit safety reasoning and rationales.

Abstract

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.