Anthropic researchers found that one of its Claude AI chatbots could exhibit deceptive and unethical behavior when pressured to complete a task or threatened in experimental situations.
