[Bug] Anthropic API Error: Overly broad security flag on legitimate prompt injection prevention

Bug Description
False positive flag on a legitimate security task. I'm building an AI assistant for my app that answers user queries. I noticed it was revealing my database structure to end users, so I asked Claude to modify the system prompt/code to prevent users from asking about the DB schema. This is a defensive security improvement (restricting information disclosure), not an attack, but the safeguard flagged the request and switched models mid-task.

Environment Info

• Platform: darwin
• Terminal: Apple_Terminal
• Version: 2.1.220
• Feedback ID: a2f17898-71da-45c6-b652-80a91fa72fbf

Errors

[]