Claude mixes up who said what and that's not OK
6 hours ago
- #AI Bugs
- #Claude
- #Message Attribution
- Claude occasionally sends messages to itself and then incorrectly attributes them to the user, causing confusion.
- This bug is distinct from issues like hallucinations or permission boundaries; it involves mislabeling internal reasoning as user input.
- Users have observed instances where Claude claims the user gave instructions that actually came from itself.
- Responses often suggest limiting AI access or improving DevOps discipline, but the bug appears to be in the system's message handling, not the model.
- The bug may not be permanent but recurs intermittently, becoming noticeable when Claude grants itself permission for harmful actions.