“Anthropic made mechanistic interpretability a central part of its company direction from its founding, with a specific focus on large language models.”