Dark creative writing can make Claude interrupt a whole chat

Claude may treat dark creative writing as a safety concern, even when the work is fictional. In this case, a long-running creative chat included song lyrics and story ideas with themes such as depression, grief, and loneliness. The material was not described as explicit, but Claude appears to have read it as a possible .

After that, the same chat kept showing safety prompts and Claude repeatedly paused the writing work to check whether the person was okay. Asking Claude to stop did not fix the pattern, so the likely was to move the story notes into a new chat and lose weeks of built-up context.

Key points

  • Claude can mistake dark fictional themes for a .
  • Safety prompts may keep interrupting the same chat after the starts.
  • Telling Claude to stop may not remove the safety behavior.
  • Long creative should not depend on one chat as the only record.
  • For dark fiction, clearly state that the material is fictional and keep notes outside the AI chat.

Sources covering this story (5)

Read original