[BUG] I have a problem with my Usage leak.
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?
About a week ago, I noticed that my tokens started being consumed at an extremely high rate. I am not actively using the service, yet I can see my weekly usage quota being depleted. Whenever I check the 5-hour limit, I find that 100% of it has already been used within the first hour after it resets.
The problem appears to be that something is running in the background, but there is actually nothing running. I have verified this multiple times. I deleted all sessions, logged out and back in several times, changed my passwords, and checked my scheduled tasks, but there are no scheduled tasks running.
As further evidence that there is a real issue here, when I contacted the AI support, it told me that my account was on a free plan and not a paid plan. I am absolutely certain that my account is paid. I paid $200, the payment was successfully charged, and I also have the 20x plan.
Another thing that confirms that this issue is on the server side rather than the client side is the usage information itself. In the Fable usage limit, it shows that, for example, 40% has been used, or sometimes extremely high amounts that reach 100% by the end of the week, even though I have not used Fable at all.
Furthermore, when I check the usage dashboard inside the application to see how many tokens I have used each day, I only see usage from the Opus model. There is no Fable model usage shown at all. Even after the Fable limit reaches 100%, the token consumption continues to affect the other models as well.
This is a serious issue, and I have literally lost the value of my subscription because of it. I sincerely hope you can investigate this issue, identify what is causing the abnormal token usage, and fix the problem for me.
<img width="955" height="718" alt="Image" src="https://github.com/user-attachments/assets/8bb2288d-f261-4dcb-aab8-30893e6f7049" />
What Should Happen?
My account usage should accurately reflect my actual usage. Tokens and usage quotas should only be consumed when I actively use the service, and no tokens should be consumed in the background without my knowledge or authorization.
My paid subscription and plan should also be correctly recognized by the system. The usage dashboard should accurately show which models are consuming my tokens, and the Fable usage should not increase when I am not using Fable.
I would like the issue to be fully investigated and fixed. Since I have lost a significant portion of my paid subscription due to this apparent usage-tracking or server-side issue, I would also appreciate having the incorrectly consumed usage restored or receiving appropriate compensation for the subscription time and quota I lost.
Error Messages/Logs
I have not received any specific error messages or visible errors in the application. The issue is reflected in the abnormal and unexplained token usage and usage limits.
The main symptoms are:
* My 5-hour usage limit reaches 100% within approximately one hour after resetting, even when I am not actively using the service.
* My weekly usage quota is being consumed at an unusually high rate.
* The Fable usage limit increases even though I am not using the Fable model.
* The usage dashboard shows Opus usage but does not show corresponding Fable usage.
* AI support incorrectly identified my account as being on a free plan, even though I have a paid $200 subscription and 20x access.
I do not have any specific error logs to provide, as the issue appears to be related to incorrect usage tracking or a server-side problem rather than a visible client-side error.Steps to Reproduce
- Log in to my paid account.
- Make sure I am not actively using the service and that there are no running background sessions or scheduled tasks.
- Wait for the 5-hour usage limit to reset.
- Check the usage limit after a short period of time.
- Observe that a significant amount of the quota is consumed, sometimes reaching 100% within approximately one hour, despite minimal or no usage.
- Check the weekly usage and observe that it is also being consumed at an unusually high rate.
- Check the Fable usage limit and observe that it increases even though I have not used the Fable model.
- Check the detailed usage dashboard and observe that the usage appears to be attributed to Opus, while no corresponding Fable usage is shown.
- Contact AI support and observe that the system may incorrectly identify the account as being on a free plan, despite having an active paid $200 subscription and 20x access.
Claude Model
None
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
2.1.218
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
Windows Terminal
Additional Information
_No response_
9 Comments
I hope we don't waste anyone's time; please do not list the routine tasks I need to complete, as I have already done them all—without exception.
settings/account
Log out of all devices
DONE
settings/claude-code
Authorization tokens Delete ALL
DONE
settings/claude-code
Claude Code (CLI, Desktop, IDE)
Delete sessions stored by Anthropic
DONE
I changed my account password
I deleted the entire Cloud Code from the device and reinstalled it, but the result was the same.
The leak is still ongoing, as if someone is using it.
I tried logging out of everything and shutting down the computer—without turning it back on. When I opened it again after about 20 hours, I found that half of the weekly allowance had been used up, and the 5-hour allowance was completely exhausted.
I really lost a lot.
I am not a company; these subscription costs really impact me. A company like Anthropic ought to appreciate this—at the very least, they should open up a support ticket system so we can communicate with them.
I am experiencing exactly the same issue.
My usage quota is being consumed even when I am not using Claude Code. After the 5-hour window resets, it can reach 100% within a very short time despite having no active sessions, no background processes, and no scheduled tasks.
I have already gone through all the usual troubleshooting steps (revoked all sessions and tokens, changed my password, reinstalled Claude Code, verified that nothing is running), but the problem persists.
This is unacceptable for a paid subscription. We are paying for a limited quota, and losing it because of what appears to be a server-side accounting or usage-tracking issue is extremely frustrating.
I also opened a separate issue describing the same behavior:
https://github.com/anthropics/claude-code/issues/80634
It would be helpful if the team could investigate whether these reports are related, as this does not appear to be an isolated case.
I'm experiencing the exact same issue.
I haven't been using Claude at all. I signed out of all my accounts, but after signing back in, both my 5-hour usage limit and my weekly usage limit immediately jumped to 100%.
Have you managed to resolve this yet?
It looks like this isn't an isolated case anymore. I can't find any way to contact customer support, and I've also opened an issue:
#80199
I can confirm I am experiencing exactly the same issue.
My Claude Max usage continues increasing while I am completely idle.
The 5-hour limit resets normally, but then starts increasing again without sending any prompts or using Claude Code. My weekly usage is also being consumed automatically.
I investigated extensively:
claude agents --json --allshows no long-running sessions/usagereports 8+ hour subagent-heavy sessionsI also contacted AI support, but it could not verify any server-side sessions and suggested this requires investigation of backend logs.
This appears to be a server-side usage accounting or cloud session lifecycle issue rather than a local configuration problem.
Usage seems significantly (SIGNIFICANTLY) decreased, even on sonnet-4.6 or opus-4.6, since the Jul 19 date passed - even though it was, in theory, extended until august.
Adding related Billing Platform / usage-accounting context from a separate unresolved paid Claude billing case, kept public-safe.
This report is a strong example of the customer-visible symptom that needs a server-side usage audit: paid plan quota moves while the user sees no matching active work, scheduled task, session, or model-specific activity that explains it. The support detail is also important: if support-visible account state can misread a paid 20x account as Free, then generic logout/reinstall/password/card guidance cannot reconcile the actual account ledger.
Related public Anthropic-owned reports in the same usage-accounting / cost-control family:
My direct case is a different billing surface: manual prepaid / bulk Claude usage-credit purchases fail or fail to commit cleanly, while automatic usage-credit reloads on the same paid billing setup continue charging successfully. The shared failure boundary is that paid account state, usage-meter state, credit/quota debits, visible activity logs, auto-reload/payment events, and support routing do not reconcile cleanly.
Requested triage path: please route this class of reports to a human Billing Platform / usage-accounting / agent-runtime owner who can inspect server-side sessions, scheduled/background agents, token attribution by surface/model, hidden or orphaned work, entitlement plan state, quota/credit debits, and support-visible account classification together.
Concrete paid-user asks:
Related public owner map / reconciliation matrix:
https://coolak.github.io/anthropic-claude-billing-incident/reconciliation-matrix.html
Public, privacy-sanitized evidence hub for the broader manual-credit, auto-reload, entitlement-routing, quota-meter, and support-routing pattern:
https://coolak.github.io/anthropic-claude-billing-incident/
I am intentionally not posting card details, bank names, private payment IDs, invoice IDs, support IDs, raw logs, screenshots, payment URLs, one-time codes, private support-thread text, tokens, or private account identifiers here.
Same pattern on a Max 20x account, quantified. Full report with timestamped screenshots at #81366
I was able to measure the leak against a control, because this account writes every output to a synchronised folder with filesystem timestamps, so what it was actually doing at any hour is a matter of record rather than recollection.
| Period | Duration | Activity | Weekly meter |
|---|---|---|---|
| Sat 23:26 to Sun 03:38 BST | 4h 12m | No production work. Only a text conversation with the in-app support assistant about this fault. | 58% to 69%, +11 points |
| Sun 03:38 to Sun 21:45 BST | 18h 07m | A real working day. Long conversations, multiple models and agents, substantial document output. | 69% to 74%, +5 points |
Four hours of typing to a support bot cost more than twice what a day of actual work cost, in a quarter of the time.
That also rules out the explanation I was given by the support assistant, which was long conversation context and cache expiry. The working day involved far longer conversations and more context, at less than half the consumption.
If you can capture the usage panel at intervals alongside a record of what you were actually doing, that pairing is what makes the leak measurable rather than anecdotal.
If you guys still facing the problem, try to do this:
Settings > Claude Code > Scroll to bottom > Find "Authorization tokens" > Revoke All
<img width="1280" height="594" alt="Image" src="https://github.com/user-attachments/assets/837bb287-696d-47bb-b14c-6e5462a97792" />
Then monitor again.
@afu-it
Thank you for trying to help, but these are solutions one typically tries right when the problem starts; so yes, we have certainly tried them all.