[BUG] Possible Usage Credits Re-enabling and Runaway Usage/Billing After Claude Code Update and Commit Action
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?
On July 15, 2026, I experienced a severe unexpected usage and billing incident while using the official Claude Code GUI.
Before the incident, Usage Credits were disabled in my account.
The observed sequence was:
- I installed a Claude Code update.
- I was using the Opus model and performed a single Git commit action through the Claude Code GUI as part of my normal development workflow.
- Shortly afterward, Claude Code usage began increasing extremely rapidly.
- My usage limits were exhausted within minutes.
- Fable usage appeared even though I had not intentionally selected or used the Fable model.
- Usage Credits appeared to become active despite previously being disabled.
- I attempted multiple times to disable Usage Credits, but the setting appeared to become enabled again before I was eventually able to turn it off.
- I manually reduced the spending limit during the incident in an attempt to contain the unexpected charges, first to $100, then $20, and eventually $1.
- Multiple unexpected billing events were generated, totaling $382.54.
The Usage Credits dashboard later showed approximately $263.61 in usage credits spent, and the account balance showed -$0.01.
I did not intentionally initiate a long-running agent task, select the Fable model, enable Usage Credits, or authorize the resulting abnormal usage.
The issue appears to involve a possible combination of:
- unexpected runaway usage after a commit action;
- Usage Credits state not consistently persisting after being disabled;
- unexpected model attribution;
- unexpected plan/billing state changes.
I cannot determine the exact technical cause and am not claiming a definitive root cause. However, the sequence appears reproducible as a potential software or account-state anomaly and should be investigated.
The incident caused a significant financial impact. The total unexpected charges were $382.54, of which $247.08 remains disputed after a separate prorated subscription refund.
Relevant context:
- Product: Claude Code official GUI
- Date: July 15, 2026
- Model intentionally used: Opus
- Model unexpectedly appearing in usage: Fable
- Usage Credits: disabled before incident; repeatedly attempted to disable during incident
- Unexpected usage: approximately $263.61
- Total charges: $382.54
- Remaining disputed amount: $247.08
I can provide screenshots and additional private evidence to the appropriate team if required.
The version i was using was most probable is Claude Code 2.1.208, i guess while facing the issue billing.
What Should Happen?
When a user performs a normal Git commit action from the Claude Code GUI, Claude Code should only perform the requested commit operation and should not unexpectedly start or continue an unbounded agent process, rapidly consume usage limits, switch models, enable Usage Credits, change subscription or billing settings, or generate unexpected API charges.
Usage Credits should remain disabled unless the user explicitly enables them, and a disabled setting should not become enabled again without a clear, intentional user action.
If a commit-related process encounters an error, loop, or unexpectedly high usage, Claude Code should stop or enforce appropriate runaway-loop and usage safeguards rather than continuing to consume usage and generate substantial charges without clear user confirmation.
In my case, the behavior I observed was the opposite: after updating Claude Code and performing a single commit action, usage increased extremely rapidly, usage limits were exhausted within minutes, Fable usage appeared even though I had intentionally been using Opus, and Usage Credits appeared to become active despite having previously been disabled.
Error Messages/Logs
Steps to Reproduce
The Claude Code version involved was most likely 2.1.208, although I cannot confirm the exact version with absolute certainty. The incident occurred around the time I updated Claude Code and then performed a commit action through the GUI. I later became aware that subsequent Claude Code releases, including 2.1.211 and 2.1.212, included changes related to billing behavior and runaway-loop limits. I am providing this version information as approximate context and not claiming that a specific release was definitively the root cause.
Claude Model
Opus
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
2.1.210
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
WSL (Windows Subsystem for Linux)
Additional Information
Relevant context:
- Product: Claude Code official GUI
- Date: July 15, 2026
- Model intentionally used: Opus
- Model unexpectedly appearing in usage: Fable
- Usage Credits: disabled before incident; repeatedly attempted to disable during incident
- Unexpected usage: approximately $263.61
- Total charges: $382.54
- Remaining disputed amount: $247.08
<img width="941" height="714" alt="Image" src="https://github.com/user-attachments/assets/2a614271-888f-444d-a434-b8af96c014e3" />
<img width="957" height="711" alt="Image" src="https://github.com/user-attachments/assets/e83cb576-3d46-4325-bbf5-3a89d0618b03" />
<img width="540" height="677" alt="Image" src="https://github.com/user-attachments/assets/181e3bc1-d5b2-4b9c-ac06-47b259c07112" />
<img width="565" height="640" alt="Image" src="https://github.com/user-attachments/assets/af7a3a4c-1552-4e8e-925c-73152c4862ab" />
<img width="581" height="657" alt="Image" src="https://github.com/user-attachments/assets/cd01710e-ab82-4a8c-b8bc-49a05189d38b" />
<img width="554" height="652" alt="Image" src="https://github.com/user-attachments/assets/fefc40f9-cbfa-4494-a991-a06156995eba" />
<img width="533" height="655" alt="Image" src="https://github.com/user-attachments/assets/610b9836-7271-4cde-8bcb-8c89579f77b3" />
<img width="538" height="673" alt="Image" src="https://github.com/user-attachments/assets/70189cf6-9288-429d-8d0a-2b20d11a6201" />
<img width="541" height="624" alt="Image" src="https://github.com/user-attachments/assets/a769ee8b-189a-4a20-a697-a71a29af4ae2" />
<img width="554" height="417" alt="Image" src="https://github.com/user-attachments/assets/01ceffd7-e254-4323-9ce9-c7159d16cbc7" />
<img width="965" height="730" alt="Image" src="https://github.com/user-attachments/assets/1e56281e-e897-4c67-8b9f-0c006b618206" />
3 Comments
Seconding this. Noticed this too
Adding related public context because this is another usage-credit control / billing-ledger reconciliation problem: account controls appear to say one thing, while runtime behavior and credit consumption can move through another path.
My current case is a different surface, but it points at the same ownership gap:
Public, privacy-sanitized evidence hub:
https://coolak.github.io/anthropic-claude-billing-incident/
Billing reconciliation matrix / owner map:
https://coolak.github.io/anthropic-claude-billing-incident/reconciliation-matrix.html
I am intentionally not posting card digits, bank names, payment IDs, invoice IDs, support IDs, screenshots, raw network logs, private support-thread text, or other private account data here. The useful ask is for Anthropic Billing Platform / payment operations to inspect usage-credit enablement state, account controls, spend limits, runtime billing-mode selection, purchase/commit state, and credit-ledger provisioning together instead of treating these as isolated user-side or generic card/browser issues.
The missing evidence here is a local run timeline for the narrow July 15 incident window. A useful reconstruction would:
That can show whether a runaway request, hidden fallback, or background branch preceded the charge. It cannot prove that Anthropic re-enabled Usage Credits or reconcile the internal billing ledger; that part still requires Anthropic's billing telemetry.
I maintain a browser-local Claude Code audit that performs that request/model/sidechain reconstruction from a selected
.claude/projectsfolder without uploading prompts or logs: https://mailcheck.agentcartai.com/tools/agent-cost-auditor/?platform=claude-code&utm_source=github&utm_medium=issue_comment&utm_campaign=claude_code_runaway_billing_78772The analysis is free. A private support-ready evidence pack is optional and costs 1 native USDC on Base Mainnet. No account credentials or raw logs are requested.