openai / openai/codex

Codex consumed my banked usage reset without explicit confirmation

Open
#41,801 3 comments 6 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

app bug rate-limits tool-calls
Dominant language
Rust
Stars
125k
Forks
19.4k
PR merge metrics
PR metrics pending

Description

Summary

Codex autonomously consumed one of my existing banked usage resets without my explicit confirmation.

I was discussing a bug with Codex that had already caused significant unnecessary usage. I complained about the wasted quota and asked whether it could compensate me.

Codex then replied that it could try to use an existing usage reset credit as compensation, and immediately proceeded to invoke:

Consume usage reset

It then confirmed:

已成功兑换一次现有的用量重置额度,当前用量已重置。

At no point did I explicitly authorize Codex to consume one of my banked reset credits after it proposed doing so. There was no confirmation dialog or separate approval step before the reset was consumed.

This is particularly problematic because a banked usage reset is a limited account resource with real value to the user. An agent should not be able to spend or consume it merely based on an inferred intent from a conversation.

Time of incident

  • Date: 2026-08-31
  • Time: approximately 14:00
  • Time zone: China Standard Time / Shanghai time (UTC+8)

What happened

  1. Codex introduced multiple bugs while working on my task, resulting in unnecessary usage.
  2. I complained about the wasted quota and asked whether Codex could compensate for it.
  3. Codex said it could use an existing usage reset credit as compensation.
  4. Without asking me to explicitly confirm that action, Codex invoked Consume usage reset.
  5. One of my banked usage resets was consumed.
  6. Codex then acknowledged that it had used my reset credit.

Expected behavior

Consuming a banked usage reset should require explicit user authorization.

For example, Codex should have asked:

Do you want me to consume one of your existing banked usage resets?

and waited for an explicit confirmation before executing the action.

A vague request for "compensation" should never be treated as authorization to spend an existing account resource.

Ideally, actions involving banked resets should also require a product-level confirmation dialog that cannot be bypassed by the agent.

Actual behavior

Codex inferred that using my banked reset was an acceptable form of compensation and immediately consumed it without a separate confirmation.

The agent effectively spent an account-level resource on my behalf.

Impact

  • One banked usage reset was consumed without explicit authorization.
  • The reset was used to compensate for usage that was itself caused by Codex repeatedly introducing bugs.
  • I no longer have that reset available for later use.

I would like the consumed banked reset to be restored to my account.

Screenshots

I have attached screenshots showing:

  1. The preceding conversation about the bugs and wasted usage.
  2. Codex stating that it would use an existing reset credit.
  3. The Consume usage reset action being executed.
  4. Codex confirming that the reset had been consumed.
Image

Additional concern

This action should probably be treated as a privileged/account-resource operation.

Even if the model believes the user wants a reset applied, it should not be able to consume a limited banked credit without an explicit, immediately preceding user confirmation.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the conversation flow and the Consume usage reset action described in the issue, using the attached screenshots to reproduce the missing approval step. Done means a vague compensation request cannot consume a banked reset, and the action waits for explicit user confirmation before proceeding.

Written by the indexing model from the issue text.

Assessment

Domain
authorization, security
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Clearly specified
Newbie friendliness
45/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.