Correction, October 13, 2026: The comparison concerns Fable 5.1’s published consumer-app prompt, not a universal API rule. Its drawing example explicitly declines the requested character before offering an original.

Anthropic tightened copyright instructions in Claude Fable 5.1’s published consumer-app prompt. It directs Claude to refuse protected lyrics and book passages. For code-based drawings, it directs Claude to decline the requested character and offer an unrelated original.

Simon Willison spotted both changes by comparing two versions of Claude’s system prompt. That prompt contains instructions Anthropic gives Claude in its consumer applications, including Claude.ai. It does not establish the same behavior for every API integration. Anthropic publishes current and older versions, so the changes can be checked directly.

What changed

Willison focused on the timing. “I doubt it’s a coincidence that they added this section within days of the news breaking that Sony Music Publishing and Warner Chappell are suing Anthropic for training on databases of song lyrics!” he wrote.

The lawsuit was filed on August 28, 2026, in California federal court. It accuses Anthropic of training Claude on pirated and scraped song lyrics. It also alleges that Claude can reproduce copyrighted lyrics word for word.

Days later, the refusal language appeared in Claude’s public system prompt. The rule includes an exception for song lyrics and poems first published before 1929. That is the prompt’s stated cutoff, rather than a complete guide to copyright status.

Willison offered a separate explanation for the drawing change. “I imagine Anthropic hadn’t worried about this before since, unlike OpenAI and Gemini, they don’t offer a specialist text-to-image model.” Claude’s recent code-based drawing ability may have made that risk more important.

Who Willison is

Willison is an independent open source developer. He created Datasette and co-created the Django web framework. His blog tracks how AI models and coding tools behave in practice.

Timing suggests a connection, but does not prove it

Willison’s comparison is strong because it uses Anthropic’s own published text. Anyone can inspect the added language. The timing is also notable. A lawsuit about song lyrics came just before a rule specifically blocking song lyrics.

The timing does not prove causation. Anthropic has not confirmed that the lawsuit caused the change. Willison’s proposed connection remains an inference from the timing.

Why builders should care

If your workflow uses Fable 5.1 through Claude’s consumer apps, reproduction requests may receive a firm refusal. API-based study apps or lyrics tools should check their own configured model and instructions.

If Claude generates code that draws characters or logos, test the output. The published example names and declines the requested character before offering an original design. Check both the explanation and the generated design against the user’s request.

Anthropic’s public prompt history is also a useful monitoring tool. Builders can check it for behavior changes before those changes surprise users in production.

End of article