2025/09/11 by Jiaqi Weng, Weng, Jiaqi, Han Zheng +14 · 1 citation
Computer Science · #Autoencoder #Feature (linguistics) #Interpretability #Interpretation (philosophy) #Limiting #Set (abstract data type) #Software Reliability and Analysis Research #Software deployment #Topic Modeling
paper · pdf · doi:10.18653/v1/2026.findings-acl.944
openalex created_date 2025/10/03 · openalex publication_date 2026/01/01 · openalex updated_date 2026/08/05
Sparse autoencoders (SAEs) enable interpretability research by decomposing entangled model activations into monosemantic features.However, under what circumstances SAEs derive most fine-grained latent features for safety-a low-frequency concept domain-remains unexplored.Two key challenges exist: identifying SAEs with the greatest potential for generating safety domain-specific features, and the prohibitively high cost of detailed feature explanation.In this paper, we propose Safe-SAIL, a unified framework for interpreting SAE features in safety-critical domains to advance mechanistic understanding of large language models.Safe-SAIL introduces a pre-explanation evaluation metric to efficiently identify SAEs with strong safety domain-specific interpretability, and reduces interpretation cost by 55% through a segmentlevel simulation strategy.Building on Safe-SAIL, we train a comprehensive suite of SAEs with human-readable explanations and systematic evaluations for 1,758 safety-related features spanning four domains: pornography, politics, violence, and terror.Using this resource, we conduct empirical analyses and provide insights on the effectiveness of Safe-SAIL for risk feature identification and how safetycritical entities and concepts are encoded across model layers.All models, explanations, and tools are publicly released in our open-source toolkit 1 and companion product 2 .