[BUG] Claude Code with Opus 4.7 just started having a conversation with itself, and began unilaterally taking action on my workstation #52228

Preflight Checklist

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

What's Wrong?

I tasked Claude Code with monitoring a data download into my database. The monitor started by showing expected status. However, after a while, it started posting a full blown dialogue between Claude and "Human:". The latter began posting questions and statements from past sessions or old documents in my project archives. Before long, Claude was providing lengthy answers and triggering pointed responses from "Human:". I had stepped away briefly. When I returned and saw this, I panicked and stopped the monitoring operation. I asked Claude what was going on and showed it a copy of some of the interactions. Claude responded that "those prompts were you". Obviously they weren't. This was incredibly disconcerting, and I see it as a massive risk and a huge flaw. I never experienced anything like this before Opus 4.7, which is what I was using.

What Should Happen?

Someone from Anthropic should immediately investigate. You should consider possibly pulling Opus 4.7 if you can't find and fix the problem straightaway.

Error Messages/Logs

I cannot post my dialogue here because the information / interaction contains confidential information. I will be happy to share directly with Anthropic via a secure channel, or authorize access to my transcript.

Steps to Reproduce

I asked Claude Code to monitor a download process as noted above. That's all it took.

Claude Model

Opus

Is this a regression?

No, this never worked

Last Working Version

No response

Claude Code Version

2.1.118

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Other

Additional Information

No response