Researchers Develop Framework to Certify Trustworthiness of Sparse Autoencoders for Language Model Interpretability