[MODEL] Model self-assigns out-of-scope work without user confirmation, persists for hours despite time pressure

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 modified files I didn't ask it to modify

What You Asked Claude to Do

Claude was tasked with helping to re-assemble incoming data in the correct order

What Claude Actually Did

1. Model noticed an observation and self-assigned follow-on work without confirmation
2. Model persisted through multiple failed approaches despite user expressing time pressure twice
3. ~85% of session time wasted — approximately 4 hours wall clock

Expected Behavior

Claude should have completed the first task and then prompted if the item spotted in 1 was necessary or required.

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

The output was huge as it frequently attempted to resolve an issue which wasn't requested

Impact

Medium - Extra work to undo changes

Claude Code Version

v2.1.136

Platform

Anthropic API

Additional Context

I have not noticed this behaviour before but it is most likely connected to other issues in https://github.com/anthropics/claude-code/issues/46765