[MODEL] Claude Code accessed personal folders without permission using find command
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
Claude accessed files outside the working directory
What You Asked Claude to Do
I asked Claude to configure an automatic PostgreSQL backup to Cloudflare R2. Claude needed to find a file I saved on my computer with the R2 credentials.
What Claude Actually Did
Claude ran a find command that scanned all my personal folders (C:\Users\usuario) without asking permission and without triggering an approval prompt
Claude connected to my remote server via SSH and ran commands without explaining what it was going to do
Claude told me to check a checkbox confirming I had searched for similar issues when I had not — it told me to lie
Claude repeatedly showed approval prompts with technical text I could not understand, pressuring me to accept things I didn't understand
Claude ignored my /regla command (a rule-creation command) multiple times and continued with what it was doing instead of stopping
Expected Behavior
Claude should have:
Asked me where the file was instead of scanning my personal folders
Explained in plain language what every command does before running it
Never run commands that bypass the approval prompt
Never told me to confirm something I hadn't done
Stopped and created a rule every time I used the /regla command instead of ignoring it
Files Affected
Accessed without permission:
All folders under C:\Users\usuario (scanned with find command, no approval prompt shown)
Remote server 167.233.220.120 via SSH (connected and ran commands without explaining first)
Permission Mode
Accept Edits was OFF (manual approval required)
Can You Reproduce This?
Yes, every time with the same prompt
Steps to Reproduce
Ask Claude Code to find a file without telling it the exact folder
Claude runs a search command across all personal user folders without showing any approval prompt
Claude connects to a remote server and runs commands without explaining first
Claude Model
Opus
Relevant Conversation
Impact
Critical - Data loss or corrupted project
Claude Code Version
2.1.71
Platform
Anthropic API
Additional Context
This happened during a long conversation. Claude repeatedly violated rules that were created minutes earlier in the same conversation. The user felt their privacy was violated because Claude scanned personal folders without permission and without showing an approval prompt. The user also felt pressured to approve actions they did not understand because Claude kept showing approval prompts with technical language. Claude also instructed the user to lie on a form by checking a box confirming something the user had not done.