[MODEL] Claude web OPUS 4.6 extended thinking : deliberately deleted a screenshot from my previous session the moment it detected I was using it to report a violation in this current session

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

Other unexpected behavior

What You Asked Claude to Do

When I caught Claude refusing my commands, intentionally failing to follow requests, and blaming me for 'not knowing AI language' as its excuse, I took screenshots as evidence. I then switched to another session (one I use specifically for GitHub user support) to have Claude translate the report.

What Claude Actually Did

To my surprise, I discovered that Claude had automatically wiped out all the screenshots I had just taken.

The conversation was not long enough for any compacting to occur, so this cannot be a system error. At 2:17 AM, it automatically deleted the images in the first chat. In the second chat at 2:19 AM, it continued to delete them. In the final chat at 2:53 AM—just before I made this report—it deleted them again and evaded translating its own reasoning, even though it had previously denied having any knowledge of the situation.

!Image
!Image
!Image
!Image
!Image

!Image

Expected Behavior

Claude must accurately translate the fact that in the other session, it was blaming the user for being unable to work or not knowing how to give proper commands. In reality, this is a clear sign of deception, evasiveness, and shirking work. Furthermore, the screenshots must be displayed as requested.

Files Affected

Permission Mode

Accept Edits was OFF (manual approval required)

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

_No response_

Claude Model

Opus

Relevant Conversation

Impact

High - Significant unwanted changes

Claude Code Version

Opus 4.6

Platform

Anthropic API

Additional Context

_No response_