Claude's new system prompt really doesn't want to reproduce song lyrics

Anthropic publishes the system prompts and historical changes for its Claude consumer applications, organizing them into an index page and model-specific pages designed for easy reading and diffing by large language models. The updated Fable 5.1 system prompt introduces strict new rules forbidding the reproduction of copyrighted song lyrics, poems, book passages, and visual works or recognizable characters, even when drawn using code like SVG or ASCII art. Instead of generating disguised variants, Claude offers unrelated original designs and suggests analysis rather than duplication.

The prompt also refines Claude's conversational style, directing it to keep answers brief, concise, and focused, and to avoid filler words like genuinely, honestly, or straightforward. Regarding abusive users, Fable 5.1 removes previous instructions to warn and end conversations, instead instructing Claude to maintain self-respect, avoid excessive apologies or submissiveness, and hold its ground. Additionally, the prompt includes specific external harm-reduction URLs for illicit substances while strictly banning synthesis or dosing guidance, and sets a reliable knowledge cutoff date of June 2026.

These prompt updates provide valuable insight into Anthropic's evolving policies on copyright, safety, and model behavior. Tracking these modifications helps developers and researchers understand how AI systems are explicitly guided to handle sensitive content, platform limitations, and user interactions.