Articles in this section aim to document how Anthropic trains and evaluates the Claude language models. Map Claude’s character training Automated behavioral audit Agentic misalignment Eval-awareness Adherence to the constitution eval Model welfare assessments