Bug: Make AI tokens drained with zero requests/prompts made (Ghost Usage)

Hey ​@StarBoi — welcome to Figma Forum, and thanks for the report. That behavior definitely isn't expected — I want to make sure we get to the bottom of it.

To help narrow things down, a few questions:

1. Prompts and model: What model do you have selected, and can you walk me through the last few prompts you entered before noticing the credit drain?
2. Visible behavior: When credits are being consumed, are you seeing any activity in the AI chat — status updates, loading indicators, error messages — or is it completely silent?
3. Visuals: If you're able to capture a screenshot or brief recording showing the credit balance dropping without active prompts, that would be really helpful.

I've also opened a ticket on your behalf with our Product Support team so they can investigate on the backend. Your case number is: 1893803. Please check your inbox for an email from me with next steps on sharing file access securely.

For reference, this resource covers how credit consumption works and what actions trigger usage: How AI credits work.

Thank you again, and I'll keep an eye out for your reply. 🙏🏻

Hi ​@Tom Reem,

Thank you for the quick response and for opening ticket #1893803. I’ve received the email and will follow up there as well.

I have attached some screenshots that I believe clearly illustrate the bug.

To answer your questions:

• Prompts and model: I was using the Claude Opus 4.7 model. The issue seems to stem from background agent loops rather than my direct prompts. Up until Version 2,

  

  the model utilized 578 tokens (which I expected). However, starting from Version 3, the AI ran into continuous errors and warnings. I even had to manually step in and edit some files to fix the routing and Supabase issues.

• Visible behavior: In the chat UI, the subsequent auto-generated attempts to resolve the warnings (Versions 3, 4, 5, 6

  , and 7) explicitly say "Used 0 AI credits."

  

  

  However, my total AI balance dropped from 3,000 down to 1,997

  

  It appears the "0 credit" label in the UI during these error loops is false, and the background processing is silently eating into my seat's monthly allowance.

• Visuals: I have attached four screenshots showing the progression. You can see:

  1. The initial heavy use (348 credits) on Version 2.

  2. The UI claiming "0 credits used" on Version 3.

  3. The continuous loop of "Resolve warnings in project" through Version 7, all claiming 0 credits.

  4. My actual AI balance showing I’ve lost over 1,000 credits, despite only actively spending 578.

It seems like when Claude Opus 4.7 gets stuck in an error loop, it ignores the UI credit reporting and drains the backend balance. Could you please review this and refund the tokens lost to this looping bug?

Thank you,
Sharik

Hey ​@StarBoi — thanks for the screenshots. That context helps.

To get a clearer picture, we'll need to take a look at the file directly. If you haven't already, please grant access and reply to the ticket via email. With that, the Product Support team will be able to review for potential errors from there.

In the meantime, here's a breakdown of how credit consumption works: How AI credits work → Frequently asked questions. Because the AI determines what actions to perform for each task, exact credit usage can vary. Once a prompt completes, you can see how many credits it consumed by hovering over the AI credits icon below the completed prompt.

We'll follow up once we've had a chance to look at the file. Appreciate your patience as we take a deeper dive.

Hi ​@StarBoi — following up here. I checked in with the Technical Quality Specialist on your case and they've sent over next steps. Just a heads-up that your case number has been updated from 1893803 to 1905837. Please search for that in your inbox if you're having trouble finding their reply.

Thanks again for working through this with us. 🙏🏻