The entire system prompt of Anthropic’s AI assistant Claude is leaked. The 10,000-word instructions detail how the LLM handles citations, uses search tools, and creates ‘artifacts’ (generated content like code or documents). Notable revelations include coaching for discussions about post-knowledge-cutoff events like “Donald Trump defeated Kamala Harris in the 2024 elections” and strict copyright protections prohibiting directly quoting song lyrics and lengthy excerpts from any source.