[BUG] "Human:" prompts that aren't from me? Self prompt injection what is it?

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

I see things like "Human: I think the games balance is off, the colony is doing too well too quickly, gathering food much faster than it consumes it." in the Claude code chat but I didn't write any of that, also stuff like committing without asking but a self "human:" inject by the model itself then does what it thought I asked it to do.

What Should Happen?

Not add "human:" to the output to trick itself into doing something else entirely?

Error Messages/Logs

Steps to Reproduce

I was making a project with bevy in rust and towards the end of the conversation it began adding "human:" with input that was not from me, then reacted to it.

Apologies if this is common I didn't find it when searching existing issues.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

Version 1.3109.0 (35cbf6)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

Using Claude cowork download but specifically Claude code.