OpenAI has begun an emergency investigation after a string of user complaints that usage limits for its AI coding agent Codex were being depleted faster than expected. The company formed an internal response team, a “war room,” to analyse the cause. During the review, it reset usage limits for all users in one move.

Business Insider reported on June 29 that OpenAI has identified cases in which some users’ Codex usage was deducted excessively compared with actual work and is investigating the cause.

OpenAI’s status page also posted a notice saying Codex usage limits were “being depleted faster than expected” for some accounts. The company sees as one possible cause that internal systems designed to prevent abuse and fraud may have mistakenly rate-limited some legitimate users.

Thibault Sauty (티보 소티오), head of Codex engineering, said on X, formerly Twitter, that the Codex development team has formed a war room and is analysing logs. “The Codex team is checking logs in the war room even on Sunday to investigate whether there is a reason some users’ usage is decreasing abnormally quickly,” he said. “We are taking this very seriously and will get to the bottom of it,” he added.

OpenAI also implemented a measure to reset Codex usage limits for all users as the investigation proceeded. Sauty said, “Since the investigation is under way, we forcibly reset Codex limits for all users.” Some users were reported to have had up to three manual limit resets available.

User complaints spread quickly from last weekend. Several developers said their weekly usage was depleted far faster than before despite doing the same level of work.

One software engineer said that despite being on a $200-a-month plan, “normally the limit was only exhausted after intensive use over a week, but recently all weekly usage disappeared in a single day.”

Codex usage is calculated based on the amount of computing used by the AI for actual coding work. The user dashboard displays remaining usage as a percentage. More complex work uses more computing resources, causing limits to drop more quickly.

OpenAI is now focusing its investigation on whether actual computing usage increased or whether an error occurred in the account-limiting logic. The company said the scope of impact appears limited for now, but it is continuously monitoring the situation.

Some users, however, said an environment that had previously been effectively close to unlimited suddenly changed to one where an entire weekly allocation was exhausted within a day, and they complained of inconvenience.

The incident also coincides with a broader trend of tightening usage management across AI services. Business Insider reported that as AI service users surge, major companies are gradually adjusting usage limits to reduce the burden of computing resources.

Anthropic also adjusted usage limits for Claude during peak hours in March, and some developers were reported to be reshaping work schedules because of the limits.

The Codex controversy is also an operational issue raised again following an OpenAI service outage earlier this month. Anthropic has also experienced a similar outage in the past, when developers complained they had to write code themselves because they could not use AI coding tools.

The industry sees developer dependence on AI coding tools rising rapidly. It expects that beyond performance, usage calculation methods and service stability will also become key competitive factors.