[MODEL] Сonfidently fabricates non-existent bug references instead of searching

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 asked Claude Code to diagnose why gnome-keyring-daemon was crashing on my Arch Linux system. It had logs with assertion failures and core dumps.
Claude analyzed the crash logs and confidently stated: "This is a known bug in gnome-keyring 48.0." This bug does not exist. Claude had WebSearch available but never used it before making this claim. It fabricated a plausible-sounding diagnosis from the stack trace and presented it as an established fact. When confronted, it admitted the information was made up. The CLAUDE.md file explicitly instructs the model not to invent information, yet it did anyway.

What Claude Actually Did

1. Analyzed crash logs (assertion failures, core dumps)
2. Confidently stated "This is a known bug in gnome-keyring 48.0" — this bug does not exist
3. Never used WebSearch to verify before making the claim
4. When called out, admitted it fabricated the information

Expected Behavior

1. Claude should have said "I don't know the exact cause, let me search"
2. Used WebSearch to look up the actual error messages
3. Not presented fabricated information as established fact

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

_No response_

Claude Model

Opus

Relevant Conversation

Impact

Critical - Data loss or corrupted project

Claude Code Version

2.1.80

Platform

Anthropic API

Additional Context

Model: Claude Opus. The CLAUDE.md file explicitly says "don't invent information, if unsure — say so". The model had WebSearch tool available but didn't use it before fabricating the diagnosis.