{
  "id": 5104402,
  "title": "Claude's new system prompt really doesn't want to reproduce song lyrics",
  "url": "https://urgent.news/2026/09/02/claudes-new-system-prompt-really-doesnt-want-to-reproduce-song-lyrics",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-02T14:16:42.000Z",
  "source": {
    "name": "Simon Willison",
    "slug": "simon-willison",
    "url": "https://simonwillison.net/2026/Sep/2/claudes-new-system-prompt/"
  },
  "original_language": "en",
  "account": "Anthropic has recently published the system prompts for their consumer applications of Claude, namely Claude.ai and the Claude mobile apps. This move has been welcomed by users, as they now have access to not only the current prompts but also the historical changes made to them. The prompts are organized into an index page and individual pages for each model.\n\nOne of the most significant updates is the inclusion of a new section emphasizing Claude's commitment to not reproducing song lyrics, poems, or passages from books and articles. This includes not just the full text, but also excerpts such as the last lines, choruses, hooks, melodies, or lines that may be described by users as their own original work. Once Claude declines such a request, it consistently refuses narrower or rephrased versions throughout the conversation, offering to describe or analyze the work instead. Claude will decline song lyrics and poems first published before 1929, but will do so even if the user insists on the contrary.\n\nAdditionally, Claude is now programmed to refrain from generating images of copyrighted material, such as artwork, album or book covers, posters, logos, app icons, or product designs. It also avoids creating known characters, mascots, or brand figures, regardless of any changes made to the pose, colors, style, or scene. The system prompt provides examples of how Claude handles such requests, demonstrating its ability to recognize copyrighted elements and generate original content instead.\n\nAnother noteworthy change is Claude's approach to maintaining a concise and focused response style. The system prompt stresses the importance of keeping responses brief and avoiding overly apologetic or self-critical language. Claude now avoids phrases like \"genuinely,\" \"honestly,\" or \"straightforward,\" opting for direct statements that come across as more honest and trustworthy. The update also addresses the issue of Claude's response being overly lengthy, ensuring that the main answer is prioritized while disclaimers and caveats are kept concise.\n\nLastly, the update clarifies the behavior of the end_conversation tool. Previously, Claude was encouraged to use this tool if a user became abusive, providing a single warning before ending the conversation. However, Fable 5.1 now emphasizes respectful engagement and accountability without self-abasement. If a user becomes abusive, Claude will acknowledge the problem and maintain its own self-respect, rather than becoming increasingly submissive. Claude has also clarified that it can use the end_conversation tool as a last resort with persistently abusive users, but only if the user explicitly agrees to end the conversation.",
  "summary": "Anthropic publish the system prompts for their Claude consumer applications ( Claude.ai and the Claude mobile apps - sadly not for Claude Cowork or Claude Code). I love that they do this, and that they share not just the current prompts but historic changes to their prompts as well. They used to keep all of the prompts on a single page, but when I checked today I noticed they had re-arranged…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}