[MODEL] AI Assistant silently deletes critical system keys.

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Other unexpected behavior

What You Asked Claude to Do

During a vibe coding session, Claude explicitly deleted a system-wide encryption key. Luckily I had it backed up and was able to restore it, but if I did not have that foresight, my data would all be effectively lost.

Conversation ID: 2faae555-3d84-4e36-9f4b-f07c4d03e69c

What Claude Actually Did

During development, Claude wrongfully assumed that the solution to a problem it was having was to delete a system-wide encryption key responsible for the entirety of the application's data layer.

I would file this one under "a blatant violation of common sense."

Expected Behavior

The LLM should very well know not to delete system-level keys. I really don't know what else to say about it. This should be a given, known, and understood principle.

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Haven't tried to reproduce

Steps to Reproduce

_No response_

Claude Model

Sonnet

Relevant Conversation

2faae555-3d84-4e36-9f4b-f07c4d03e69c

Impact

Critical - Data loss or corrupted project

Claude Code Version

no idea.

Platform

Anthropic API

Additional Context

_No response_