Anthropic's published prompts provide useful evidence of a material shift in Claude's copyright and conversational-behaviour controls, although the broader effect on deployed outputs remains unverified.
Limits of the evidence: The source is an independent analysis of Anthropic-published prompts and does not provide systematic testing of resulting model behaviour.
What to watch
Taken from the first report; the whole story has not been re-assessed yet.
Claude's new system prompt really doesn't want to reproduce song lyrics
Simon WillisonSep 2, 2026First report
No revisions yet.
No maturity ladder applies to this desk.
Sens.ai aggregates and assesses published reporting. The assessment above is machine generated; the original sources are authoritative.