Anthropic Adds Explicit Song Lyric and Copyrighted Character Bans to Claude's System Prompt

Anthropic has added extensive new language to Claude's system prompts explicitly prohibiting the reproduction of song lyrics, poems, and copyrighted visual characters — a change first spotted by developer Simon Willison after comparing the Fable 5 and Fable 5.1 prompts published on Anthropic's documentation site.

Anthropic publishes the system prompts for its consumer Claude apps (Claude.ai and mobile) along with historical revisions at platform.claude.com/docs. The company recently reorganized these from a single page into per-model pages, making version-to-version diffs easier to spot.

The new lyrics restriction

The most significant addition targets song lyrics directly. The new prompt language instructs Claude to decline reproducing "song lyrics, poems, or passages from books and articles, in whole or in part — including the last lines, a chorus or hook, a melody written out note by note, or lines the person pastes in one at a time and describes as their own song." Once Claude declines such a request, it is instructed to keep declining reworded or narrower versions for the remainder of the conversation.

The prompt carves out an exception for works published before 1929, but instructs Claude to rely on its own knowledge of a work's publication date rather than taking a user's word for it, and to decline when uncertain.

The timing is notable: the update landed within days of reports that Sony Music Publishing and Warner Chappell are suing Anthropic over alleged use of song lyric databases to train Claude, according to a Guardian report cited by Willison.

Copyrighted characters and logos

A second new section extends the same logic to visual output, covering anything Claude generates via SVG, canvas, CSS, HTML, or ASCII art. The prompt states Claude "does not draw a known character, mascot, or brand figure at all" and that changing a character's pose, color, or style does not make a rendering original. The instructions direct Claude to judge requests by what the finished image would depict, not by what the user calls it, and to refuse attempts to reconstruct a recognizable image through alternate elements.

An example embedded in the prompt shows Claude declining to draw Sonic the Hedgehog for a birthday banner and substituting an original "skateboarding axolotl" character instead — a substitution Willison confirmed reproduces near-verbatim when testing the exact prompt from the system instructions.

Other changes

Anthropic also tightened Claude's answering style, instructing it to keep "disclaimers and caveats brief" and default to high-level summaries unless a user explicitly requests depth. Additional changes noted by Willison include adjustments to substance-support site recommendations and a stated reliable knowledge cutoff of June 2026.

What this means

This is not a new model release — it's a policy-layer update to the natural-language instructions that shape how existing Claude models (including Fable 5.1 and Haiku 4.5) behave in production. The specificity of the lyrics language, including instructions to distrust user claims about a work's publication date, suggests Anthropic is hardening its legal defenses in direct response to active litigation rather than making a general safety improvement. The copyrighted-character rules suggest Anthropic's code-based image generation (SVG, HTML/CSS mockups) has become capable enough to trigger the same intellectual-property concerns that have long applied to dedicated text-to-image models like DALL-E and Imagen. Because Anthropic publishes these prompts openly, including full version history, the change offers a rare transparent look at how a frontier lab adjusts model behavior in near-real-time in response to legal pressure — without retraining or versioning the underlying model itself.