Codex fixes usage-limit sync bug after 2-hour subscriber lockout

• OpenAI said a usage-metering bug put "the majority of users on a subscription" at the wrong enforced Codex limit for about two hours, according to thsottiaux's incident update.
• Users were getting hard stop messages even when their dashboards still showed reserve left, as koltregaskes' first screenshot and koltregaskes' later screenshot documented.
• OpenAI said it waived usage consumed during the bad window and restored balances to where users had left off, per thsottiaux's incident update.
• The incident landed one day after OpenAI reset Codex limits across paid plans, then acknowledged reset-propagation issues on some accounts in thsottiaux's reset thread.

OpenAI's public thread was unusually specific on the user impact: thsottiaux's follow-up said the bad enforcement hit most subscribed users, lasted about two hours, and got reversed. The screenshots were weirder than a normal quota error. koltregaskes' first post shows repeated "remote compact task" failures, while koltregaskes' second post shows a sidebar claiming 0% remaining next to a weekly meter still at 37%. The same May 16 reset thread also pointed people to the new remote connections docs, which makes the timing awkward for anyone who had just started leaning on mobile Codex sessions.

The key detail is in OpenAI's wording. In thsottiaux's initial acknowledgement, thsottiaux said usage limits were "out of sync" for some Codex users. By the fuller update, the scope had expanded to the majority of subscribed users, and the bug was described as wrong enforced limits rather than unusually heavy consumption.

That distinction matches the remediation. thsottiaux's incident update says OpenAI waived usage consumption for the affected window and restored balances to exactly where users had left them, which implies the meter state, not the underlying subscription entitlement, was the thing that drifted.

The user-visible failure had at least three pieces:

• Repeated errors saying: "Error running remote compact task: You've hit your usage limit," in koltregaskes' first screenshot.
• Automatic context compaction firing alongside those errors, again in the same screenshot.
• A contradictory quota display where the left rail showed "Weekly MANAGER 37% 23 May" while the main pop-up said "You're out of Codex messages" and reset on 18 May, in koltregaskes' later screenshot.

That is a nastier failure mode than a plain outage because the product keeps surfacing state, just not one state. The compact-task error also suggests the lockout could interrupt active long-context runs instead of only blocking brand new prompts.

On May 16, thsottiaux's earlier post said OpenAI had fixed two issues behind a roughly 48-hour degradation in GPT-5.5 capability inside Codex and would reset usage limits that evening. Hours later, thsottiaux's reset thread announced those limits had been reset across all paid plans, then added that some accounts were having reset-propagation issues.

Community replies make clear some users treated the reset like a free weekly refill. argofowl's clip shows a reserve jumping back to 24%, and jeffscottward's post asked whether other heavy /goal users had also gotten a free weekly reset.

The same reset thread also carried a product nudge: in thsottiaux's remote control post, users were told to try Codex's remote control feature through the ChatGPT app while on the go. That links to OpenAI's remote connections documentation, which describes connecting Codex to remote environments from mobile and desktop surfaces.

One user screenshot from later in the incident, mattlam_'s mobile Codex view, shows the phone client loading a Codex session while replies joked about getting back to "Codex on the train." The outage report never published a root cause. When mattlam_ and the same user in a follow-up asked what caused the bad enforcement, thsottiaux's reply answered with a joke instead of an explanation.