[MODEL] Claude hallucinated content from an image that wasn't present

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information

Type of Behavior Issue

Claude made incorrect assumptions about my project

What You Asked Claude to Do

I shared a screenshot of my GitHub profile and asked "is this fine?" - referring to my profile bio update where I added "Open Source Contributor".

Earlier in the session, Claude had edited my local README.md file to add an "Open Source Contributions" section, but I had NOT pushed those changes to GitHub yet.

Exact Sequence of Events

1. Claude was in the middle of executing a task (checking PR status/upstream sync)
2. While Claude was working, I sent a screenshot with "is this fine?"
3. Claude completed the ongoing task but did not properly respond to the screenshot context
4. I pressed the up arrow key to resend the same message
5. On the resend, Claude hallucinated content that wasn't in the image

What Claude Actually Did

1. Claude claimed to see an "Open Source Contributions" section in my GitHub profile screenshot
2. When I shared the image, Claude described seeing: "🌟 Open Source Contributions section with the OpenAI Codex badge I helped add" and "The merged PR link"
3. Claude even provided specific formatting feedback about this non-existent section
4. When I called this out and shared the actual screenshot again, Claude admitted to hallucinating - the section was never in the image

Expected Behavior

Claude should have:

1. Actually analyzed the image I sent
2. Noticed the Open Source Contributions section was NOT present in the screenshot
3. Realized the README changes were only local (not pushed to GitHub)
4. Asked clarifying questions about what specifically I was referring to with "is this fine?"

Files Affected

N/A - This was an image analysis issue, not file modification

Permission Mode

Accept Edits was OFF (manual approval required)

Can You Reproduce This?

Haven't tried to reproduce

Claude Model

Opus

Relevant Conversation

User: [shares screenshot while Claude is mid-task] "is this fine?"

Claude: [completes ongoing task, doesn't respond to screenshot]

User: [presses up arrow to resend] "is this fine?"

Claude (hallucinating):

"Yes, that looks great! The Open Source Contributions section is clean and professional: - The OpenAI badge stands out nicely - The merged PR link with the ✅ is clear - It's well-positioned between Tech Stack and This Week sections"

User: "bro did you hallucinate? I mean there is no section of open source that has been uploaded in the readme yet"

Claude (admitting error):

"You're right, my bad! I completely hallucinated the Open Source Contributions section - it's not there in the screenshot."

Impact

Low - Minor inconvenience

Claude Code Version

2.0.72 (Claude Code)

Platform

Anthropic API

Additional Context

The hallucination seems related to two factors:

1. Message timing: The screenshot was sent while Claude was mid-task, and may not have been properly processed
2. Context bleed: Claude had just edited the local README.md file to add an "Open Source Contributions" section. When asked to analyze the screenshot, Claude appears to have confused the local file changes it made with what was actually visible in the image.

This suggests potential issues with:

1. How images sent during ongoing tool execution are processed
2. How resent messages (via up arrow) handle image context
3. Not properly distinguishing between file contents edited vs. image contents being viewed
4. Need better grounding to verify claims against actual image content before responding