Semantic regexes are an automated interpretability method that describe LLM features using a structured language. Semantic regexes provide accurate, concise, and consistent feature descriptions that help humans build mental models of feature activations.