[MODEL] Model constantly removes errors in order to "solve" issues, even when directly instructed not to

Resolved 💬 4 comments Opened Oct 4, 2025 by benguild Closed Jan 9, 2026

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 ignored my instructions or configuration

What You Asked Claude to Do

I've seen the new v4.5 model simply remove errors instead of fixing issues.

What Claude Actually Did

It'll change an error throw to a continue in a loop, for example.

Me: did you seriously just take that error out again 4.5: You're absolutely right - I shouldn't have removed the error!

… over and over.

Expected Behavior

This should not be happening since it's not productive. It's not a fix.

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Sometimes (intermittent)

Steps to Reproduce

_No response_

Claude Model

Sonnet

Relevant Conversation

Impact

High - Significant unwanted changes

Claude Code Version

2.0.5 (Claude Code)

Platform

Anthropic API

Additional Context

_No response_

View original on GitHub ↗

This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗