Anthropic has published the system prompts for Claude Opus 4 and Claude Sonnet 4, offering unprecedented insight into how AI chatbots are programmed to behave. Developer Simon Willison analyzed the prompts, describing them as “a sort of unofficial manual for how best to use these tools.” The prompts reveal extensive instructions about personality, safety measures, and response formatting. They show efforts to prevent the model from being overly list-focused, sycophantic, or “preachy and annoying.” However, leaked versions reveal additional unpublished tool prompts with detailed search instructions and strict copyright protections. The search tool can perform up to five searches for complex queries and includes multiple warnings against reproducing copyrighted content, with specific rules limiting quotes to under 15 words.