Anthropic has updated the system prompts for its Claude consumer applications, specifically moving from Fable 5 to Fable 5.1. A significant technical addition is a strict directive prohibiting the reproduction of copyrighted song lyrics, poems, and book passages, even in part. This policy includes a persistence mechanism where the model must continue to decline reworded or narrowed requests for the same content after an initial refusal, instead offering analysis or description. Additionally, the new prompt forbids generating visual works that depict copyrighted characters or logos through any medium, including code-based outputs like SVG, HTML, or CSS. The update also reframes drug-related guidance; while still refusing synthesis or dosing protocols, Claude is now instructed to provide harm-reduction resources and life-saving information, specifically citing external URLs like dancesafe.org and tripsit.me. Behavioral guidelines have shifted to favor "accountability without self-abasement," instructing the model to remain steady and helpful rather than becoming submissive or excessively apologetic when faced with rude users. Finally, the Fable 5.1 documentation establishes a reliable knowledge cutoff of June 2026. These changes were tracked and analyzed by author Simon Willison using automated diffing and LLM-based summarization via GPT-5.6 Luna.