# sae

Sparse Autoencoders for Security Analysis

Using sparse autoencoders and mechanistic interpretability to identify and manipulate safety-relevant features.