Controllable Safety Alignment: Inference-Time Adaptation to Diverse Safety Requirements

TL;DR

Controllable Safety Alignment (CoSA) enables large language models to adapt their safety behaviors at inference time based on user-defined safety configurations, enhancing flexibility across diverse social norms without retraining.

Contribution

We introduce CoSAlign, a data-centric method for dynamic safety alignment, and develop CoSA-Score and CoSApien benchmark to evaluate and facilitate adaptable safety in LLMs.

Findings

CoSAlign significantly improves controllability over baseline methods.

The framework allows real-time safety behavior adjustments via safety configs.

Our evaluation shows increased alignment with diverse safety requirements.

Abstract

The current paradigm for safety alignment of large language models (LLMs) follows a one-size-fits-all approach: the model refuses to interact with any content deemed unsafe by the model provider. This approach lacks flexibility in the face of varying social norms across cultures and regions. In addition, users may have diverse safety needs, making a model with static safety standards too restrictive to be useful, as well as too costly to be re-aligned. We propose Controllable Safety Alignment (CoSA), a framework designed to adapt models to diverse safety requirements without re-training. Instead of aligning a fixed model, we align models to follow safety configs -- free-form natural language descriptions of the desired safety behaviors -- that are provided as part of the system prompt. To adjust model safety behavior, authorized users only need to modify such safety configs at inference time. To enable that, we propose CoSAlign, a data-centric method for aligning LLMs to easily adapt to diverse safety configs. Furthermore, we devise a novel controllability evaluation protocol that considers both helpfulness and configured safety, summarizing them into CoSA-Score, and construct CoSApien, a human-authored benchmark that consists of real-world LLM use cases with diverse safety requirements and corresponding evaluation prompts. We show that CoSAlign leads to substantial gains of controllability over strong baselines including in-context alignment. Our framework encourages better representation and adaptation to pluralistic human values in LLMs, and thereby increasing their practicality.