Anthropic says its latest research examines how Claude expresses values across model versions and languages, based on more than 300,000 anonymized conversations. In the company's description, the work builds on earlier research that found Claude expresses more than 3,000 values, including honesty and warmth, and asks how those values vary depending on the model and the language used. So far, the public source in view is Anthropic's announcement and linked research page.