Caught the real culprit behind Codex quota running out: OpenAI patched multiple vulnerabilities and performed a third full reset with compensation

According to monitoring by Beating (Dongcha), the reason behind the abnormal quota consumption of OpenAI’s programming agent Codex has been officially identified. Tibo Sottiaux, the core product lead, announced that the team has fully deployed the fix patch. In addition to resetting the quota for all users again, every user will also receive an extra reset card that can be enabled for 24 hours.

The overconsumption was not caused by a single vulnerability; rather, it resulted from the combined effect of multiple minor issues in the backend and display false positives. At the execution level, due to the system’s automated review running too frequently, it unexpectedly triggered an excessive number of sub-agent tasks. Meanwhile, the backend suggestion feature, after failing, would repeat running and frequently retry, multiplying Token usage.

At the display level, automated review was incorrectly categorized as GPT-5.4 consumption, and requests that failed or were rate-limited were also incorrectly shown as quota consumption in the front-end charts, directly triggering a widespread quota shortage for users across the internet.

Currently, the official team has synchronized and deployed hotfix patches to the billing backend, the desktop client, and the CLI terminal. In addition to refreshing the quota for all users again, going forward, only successfully executed interaction requests will be recorded in the Turn statistics chart. Although the erroneous data in historical charts cannot be changed, the updated, actual Token consumption will be significantly lower.