[BUG] Claude Code login fails due to phantom unpaid invoice

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?

NOTE: I have attempted to use the AI chatbot while logged in to my claude.ai account several times, but it keeps giving me nonsense advice that ignores facts I have clearly articulated for it. I am filing this bug here because I don't see any other recourse.

This morning I opened up my work laptop and found my claude code sessions from last week still active. I picked up where I left off on a bug fix task on Friday and claude code was slopping along merrily for several minutes.

After several minutes of slopping to itself, it suddenly told me that I wasn't authorized and that I needed to run /login. I did this, pasted the auth URL it gave me into a browser, then clicked "Authorize" at which point I see the following in the web browser:

<img width="337" height="297" alt="Image" src="https://github.com/user-attachments/assets/36965767-be0c-4b13-8e79-efe06b651c36" />

However, when I navigate to Settings > Billing I can clearly see that all of my invoices are paid. The most recent dated Feb 26 which covers the period between Feb 26, 2026 and Mar 26, 2026, is paid. I can see the corresponding credit card transaction in my bank account dated Feb 28. Today is March 10, 2026 and I have been using my account since Feb 26 with no problem.

There is no invoice to pay. All of my invoices have been paid. There is no other recourse for me than to post here hoping it gets through the inevitable LLM filter that Anthropic is slopping at users like me.

What Should Happen?

I have already paid my Feb 26 - Mar 26 invoice. I should be able to log in. I can't.

Error Messages/Logs

Steps to Reproduce

Pay your invoice on time, use claude code for a couple weeks, then be unlucky enough to get caught up in the fallout of Anthropic "engineers" dogfooding their own LLM to generate code slop for their authz endpoint.

Claude Model

None

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.71

Platform

Anthropic API

Operating System

Other Linux

Terminal/Shell

Other

Additional Information