Anthropic updates system prompts for Claude consumer apps, including Fable 5.1

Anthropic has published updated system prompts for its consumer-facing Claude models, including Fable 5.1, Haiku 4.5, and others, on its platform documentation site. The prompts are presented per-model and include historical versions, enabling direct comparison of changes over time.

A notable addition in Fable 5.1 is a section explicitly prohibiting the reproduction of song lyrics, poems, or passages from books and articles, including partial or reworded versions. Once Claude declines such a request, it will continue to refuse narrower or reworded versions for the remainder of that conversation and will offer to describe or analyze the work instead. The rule exempts works first published before 1929, such as Shakespeare’s sonnets or Keats’ odes, but relies on the model’s internal knowledge of the work’s date rather than user-provided information.

The updated prompts also introduce restrictions on generating images of copyrighted material, including code-generated art like SVGs, canvas, CSS, HTML mockups, and ASCII art. Claude will not reproduce specific artworks, album or book covers, posters, logos, app icons, product designs, or known characters. If a request clearly identifies a recognizable character or work, Claude will decline and may offer an original alternative. For example, when asked to create a birthday banner featuring a blue hedgehog, Claude declined and instead generated an SVG of a skateboarding axolotl.

Anthropic also refined the response style in Fable 5.1, emphasizing brevity and conciseness to avoid overwhelming users. Disclaimers and caveats are to be brief, with most of the response focused on the main answer. The prompts further instruct the model to avoid using modifiers like 'genuinely,' 'honestly,' or 'straightforward,' as these are considered disingenuous.

The system prompts for Fable 5.1 include updated guidance on handling requests related to illegal substances. While the model will not provide synthesis, production, or distribution protocols, it may offer life-saving information such as dangerous interactions, overdose signs, or when to seek help. The prompts now explicitly direct users to harm-reduction resources like dancesafe.org, tripsit.me, and psychonautwiki.org.

The documentation for Fable 5.1 also specifies a reliable knowledge cutoff of June 2026, which is directly included in the system prompt as a macro: {{currentDateTime}}. This is intended to help users understand the model’s limitations and the context of its responses.