The field of transformer research is moving towards improving interpretability and robustness of these models. Recent developments focus on developing techniques for extracting explicit algorithms from neural networks, modifying specific behaviors without retraining, and evaluating the robustness of these models. Mechanistic interpretability and model editing are two areas that are being explored, with a focus on providing explicit guarantees about the behavior of the extracted or edited models. The use of certified blockwise extraction and progressive localization are showing promising results in achieving high performance while providing interpretable attention patterns. Additionally, the extraction of robust register automata from neural networks is enabling principled robustness evaluation and bridging the gap between neural network interpretability and formal reasoning. Noteworthy papers include: BlockCert, which introduces a framework for certified blockwise extraction of transformer mechanisms, and Progressive Localisation in Localist LLMs, which demonstrates the optimal architecture for creating interpretable large language models while preserving performance. Extracting Robust Register Automata from Neural Networks over Data Sequences is also a significant contribution, providing a framework for robust DRA extraction from black-box models.