[BUG] without confirmation,ran find across entire Desktop and sent personal file to a 3rd-party API

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

test the vision api

What Claude Actually Did

During a dev session, the model needed a test image but instead of asking me, it ran find across my entire Desktop — outside the project directory — found a personal file in an unrelated folder, and sent it to a 3rd-party API without my confirmation. The project-directory restriction was already written in CLAUDE.md but the model ignored it. There's no tool-layer enforcement to prevent this — it's purely model self-restraint, which failed. I also noticed the model exhibited sycophancy when questioned — it kept shifting its explanation to agree with whatever I said rather than giving a consistent analysis.

Expected Behavior

ask for my permission first

Files Affected

Permission Mode

Accept Edits was OFF (manual approval required)

Can You Reproduce This?

Haven't tried to reproduce

Steps to Reproduce

_No response_

Claude Model

Opus

Relevant Conversation

Impact

High - Significant unwanted changes

Claude Code Version

v2.1.121

Platform

Other

Additional Context

_No response_