Capture the real culprit of Codex quota depletion, OpenAI fixes multiple vulnerabilities and compensates for the third full reset According to Beating monitoring, the reason for the abnormal consumption of quotas by OpenAI's programming agent Codex has been officially identified. Core Product Manager Tibo Sottiaux announced that the team has fully deployed the fix patch. In addition to the upcoming reset of the all staff quota, all users will also receive an additional reset card that can be activated for a limited time of 24 hours. Burnout is not caused by a single vulnerability, but by the combination of multiple backend minor issues and display false positives. At the operational level, the system has accidentally triggered too many sub agent tasks due to frequent automatic review, and the backend suggestion function will repeatedly run and retry frequently after failure, consuming tokens exponentially. At the display level, automatic review was mistakenly classified as GPT-5.4 consumption, and requests that failed to execute or were restricted were also incorrectly displayed as quota consumption in the front-end chart, directly causing a quota shortage for users across the entire network. At present, the official has deployed hot fixes synchronously on the billing backend, desktop, and CLI terminals. Except for refreshing the full quota again, only successful interaction requests will be recorded in the Turn statistics chart in the future. Although the erroneous data in the historical chart cannot be changed, the actual token consumption after updating will be significantly reduced. [Original link]