Codex fixes quota drain tied to fraud overflagging with an account-wide usage reset

• OpenAI's Thomas Sottiaux said some Codex accounts were seeing usage drain "faster than intended," and tied it to abuse and fraud checks that were overflagging some sessions Thomas Sottiaux's incident post.
• Sottiaux later said OpenAI was giving all Codex users a usage reset, with mitigations already applied while the company kept monitoring the issue Thomas Sottiaux's reset announcement.
• User reports before the reset described paid tiers burning down unexpectedly, including LLMpsycho's account complaint and a support reply where dkundel's feedback request asked for session IDs that drained.
• OpenAI's public line stayed narrow: according to Thomas Sottiaux's reset announcement, the investigation had not shown users being impacted at large, even as the company issued an account-wide reset.

The interesting part is how specific the cause got, fast. Thomas Sottiaux's incident post blamed fraud and abuse overflagging rather than model demand, dkundel's feedback request asked affected users for feedback IDs tied to draining sessions, and Thomas Sottiaux's reset announcement turned that into a platform-wide quota reset instead of case-by-case credits.

OpenAI's first public explanation was unusually concrete for a quota bug. Sottiaux wrote that Codex was "investigating issues where some accounts are seeing faster usage draining than intended" and said the team believed fraud prevention mechanisms were overflagging certain sessions.

That framing matters because it separates the incident from a simple pricing or plan change. Earlier user speculation, including LLMpsycho's account complaint, compared the new $100 Codex tier to the older $20 plan and concluded "something is wrong," but the official explanation pointed at account classification, not published rates.

OpenAI's fix was broad. Sottiaux said all Codex users would receive a usage reset "on the house," that it should appear within a few hours, and that mitigations had already been applied.

That reset was quickly echoed by other accounts, including aibuilderclub_'s reset summary, which repeated that subscribers should see fresh quota land in their accounts. The company did not describe the mitigation in public, only that monitoring was still ongoing.

The support trail suggests OpenAI had already started collecting concrete failure cases before the public reset. In reply to a user report, dkundel asked for a /feedback submission and said sessions that drained especially fast would be shared with the team.

The public complaints were short, but they were consistent:

• LLMpsycho's account complaint said the effective quota on a $100 Codex account looked like the old $20 plan with doubled rates.
• dejavucoder's reset mention referenced getting a Codex rate-limit reset during active benchmarking work.
• Matthew Berman's reset prediction treated another reset as likely before OpenAI confirmed the broader account-wide credit.

The last notable detail is the mismatch between scope and wording. OpenAI issued a reset to all Codex users, but Sottiaux also said the investigation had not shown users being impacted at large.

That left some unresolved edge questions in public. After the reset announcement, vikhyatk's reply about expiring resets asked whether the credits themselves would expire, a detail OpenAI had not addressed in the visible thread.