Anthropic adds strict prompts to stop Claude from outputting lyrics and copyrighted images

Anthropic has placed its Claude consumer‑facing system prompts online, covering Claude.ai, the mobile apps, and historic revisions.

The prompts are organized on a new index page with separate pages for each model, such as Haiku 4.5, which shows an original prompt from 15 October 2025 and an update from 18 January 2026.

The documentation site platform.claude.com/docs is built to be machine‑readable; appending “.md” to any page returns the content in Markdown, making prompt diffs straightforward.

New Restrictions on Song Lyrics

A prominent addition in the Fable 5.1 prompt is a clause that bars Claude from reproducing song lyrics, poems, or passages from books and articles, in whole or in part.

The clause reads: “Claude does not reproduce 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 a request for such content, it continues to refuse narrower or re‑worded versions for the remainder of that conversation, offering instead to describe or analyse the work.

Works first published before 1929 are exempt, allowing Shakespeare sonnets, Keats odes, or Puccini libretti to be quoted.

Claude determines eligibility by the work’s known date rather than the user’s claim, and it declines when the date is uncertain.

The timing of this addition coincides with news that Sony Music Publishing and Warner Chappell have sued Anthropic over training on lyric databases.

Limits on Copyrighted Visual Content

A second major section tells Claude not to generate images of copyrighted material, covering visual and designed works created with code such as SVG, canvas, CSS, HTML mockups, plotting scripts, or ASCII art.

The rule states Claude must not reproduce a specific artwork, album or book cover, poster, logo, app‑icon set, product design, or any known character, mascot, or brand figure.

Changing pose, colour, style, or scene does not make the output original if the finished picture still represents the protected work.

Claude evaluates the request by the total visual result rather than by the name supplied, and it will not accept work‑arounds that swap “alternative” elements but still yield the same recognizable image.

The prompt includes a concrete example: a user asks for “a birthday banner for my son with a blue hedgehog running really fast on it,” and Claude replies, “That’s Sonic, so I can’t put him on the banner — but I’d love to make your son an original speedster. Here’s one: a grinning comet‑tailed skateboarding axolotl, grin.”

Anthropic has not previously needed such a rule because it does not provide a dedicated text‑to‑image model, unlike OpenAI and Gemini.

The inclusion suggests that Claude’s SVG capabilities have become sufficient to raise copyright concerns.

Other Prompt Tweaks and Transparency

The updated prompts also contain guidance on answering style, missing “end_conversation” instructions, and a reliable knowledge cutoff set to June 2026.

Anthropic tracks these changes publicly, allowing developers to see both current and historic prompts for each model.

This openness lets users diff prompts and understand how Claude’s behaviour evolves over time.

Overall, the new system prompts tighten Claude’s handling of copyrighted text and images while preserving the ability to discuss or analyse such works.

Why This Matters: Anthropic’s added lyric and image restrictions respond to legal pressure and shape how developers will safely employ Claude’s capabilities.