Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment
Disentangled Safety Adapters decouple safety computations from base models via lightweight adapters, enabling efficient guardrails and flexible inference-time alignment with minimal overhead. They boost safety metrics by up to 53% AUC and reduce alignment tax by 8 points versus standard fine-tuning.
Published May 30, 2025arXiv ↗
Only vote on papers you've read. Sign in to vote.
Abstract
Existing paradigms for ensuring AI safety, such as guardrail models and alignment training, often compromise either inference efficiency or development flexibility. We introduce Disentangled Safety Adapters (DSA), a novel framework addressing these challenges by decoupling safety-specific computations from a task-optimized base model. DSA utilizes lightweight adapters that leverage the base model's internal representations, enabling diverse and flexible safety functionalities with minimal impact on inference cost. Empirically, DSA-based safety guardrails substantially outperform comparably sized standalone models across hate speech classification, detecting unsafe model inputs and responses, and hallucination detection with relative improvements of up to 53% in AUC. Furthermore, DSA-based safety alignment allows dynamic, inference-time adjustment of alignment strength and a fine-grained trade-off between instruction following performance and model safety. Importantly, combining the DSA safety guardrail with DSA safety alignment facilitates context-dependent alignment strength, boosting safety on StrongREJECT by 93% while maintaining 98% performance on MTBench - a total reduction in alignment tax of 8 percentage points compared to standard safety alignment fine-tuning. Overall, DSA presents a promising path towards more modular, efficient, and adaptable AI safety and alignment.