Two public statements from OpenAI in 2026 show both sides of enforcement against subscription resale. One warned people running sub2api. The other was an incident report for Codex users the same systems caught by mistake.
CaseHarry Johnston, NoResponse Research6 min read
Codex subscription resale means turning a ChatGPT plan, which includes OpenAI's coding agent Codex, into API traffic that other people pay to use. OpenAI addressed it in public twice in 2026. On 21 August, Tibo, who works on Codex and ChatGPT at OpenAI according to his X profile, said this kind of use "is not something we support" and that it "gets flagged by our fraud-prevention systems". Eight weeks earlier, OpenAI's status page had recorded those systems rate limiting real Codex accounts by mistake.
Together the two statements show the cost every platform weighs before it turns on a control against resale. The systems that catch a shared plan also catch real customers, and OpenAI had to say so in public.
At a glance
The post answered complaints about "codex usage limits being different". Tibo wrote that "when talking to affected users many were using sub2api." Then Tibo wrote: "Converting a subscription into api traffic to then re-serve or share across many users is not something we support and this type of usage gets flagged by our fraud-prevention systems."
He also marked the other side. Using a subscription through Sign in with ChatGPT is "completely fine", in official clients or in open-source clients that sign in with the account.
sub2api's README calls it an "AI API Gateway Platform for Subscription Quota Distribution". One person's subscription sessions go in, and API keys for many people come out. It was created on 18 December 2025 and had more than 40,000 GitHub stars by October 2026.
On 26 June 2026, OpenAI's status page opened an incident called "Codex Usage Limits Depleting Faster Than Expected". Six hours in, the update read: "We have identified that some reports of Codex usage limits depleting faster than expected are related to our abuse and fraud prevention systems incorrectly rate limiting certain accounts." It was resolved on 29 June.
The incident does not mention resale, and we do not claim resale caused it. It shows the other face of the same control. A system tuned to catch one subscription serving many people will sometimes read a busy real user the same way.
| Angle | What a ChatGPT plan allows | What resale looks like in traffic | What OpenAI said happens |
|---|---|---|---|
| Who is served | One person, signed in with ChatGPT through an official or open-source client. Tibo: "completely fine". | One account's sessions become API keys for many people, which sub2api calls "Subscription Quota Distribution". | "not something we support", and it "gets flagged by our fraud-prevention systems". |
| How it is paid for | A flat plan with an included allowance; the Codex pricing page adds "Weekly limits may also apply." | Buyers pay per token at a discount. The seller's main upstream cost is the flat plan, plus infrastructure and payment costs. | Flagged accounts hit their limits early, reported by users as limits "being different". |
| Beyond the allowance | Extra work through an API key, "charged at standard API rates". | The relay adds more subscriptions and avoids API rates. | In June the same systems rate limited "certain accounts" by mistake. |
Tibo accepted third-party clients that sign in with the account. What he objected to was a subscription re-served to more than one person. That matches the definition of resale in our token fraud guide, where a seat licensed to one person and served to others as an API has become supply.
The upstream supply is a flat-rate plan. OpenAI's authentication docs say that with an API key, "Codex uses standard API pricing instead of included ChatGPT plan credits." Anyone who can turn plan credits into API traffic and sell it per token keeps the difference. That is the business of the transfer stations we track. Anthropic reached the same point in July 2025, when it introduced weekly limits for Claude subscribers and said a small number of users were sharing and reselling their accounts.
Some of the public complaints about Codex limits came from shared seats. Split a seat across many people and each of them sees the allowance drain fast. The June incident shows real users complaining in the same words. OpenAI separated the two groups by talking to affected users, as Tibo described, and by looking at how their accounts were used.
OpenAI's mistake ran for three days, from 26 to 29 June 2026. The incident page lists the Codex web app, the Codex API, the CLI, the VS Code extension and Codex in the ChatGPT desktop app among the affected components. Real customers lost part of their allowance to a control aimed at someone else, and OpenAI had to explain it on a public status page.
A reseller who hits a limit early buys another seat. A real customer who hits it sees a broken product and may say so in public. Three habits keep a control against resale from landing on the second group.
If you sell inference on a flat-rate plan, a complaint about limits tells you nothing on its own. Whether a seat is one developer or a storefront shows in the account's working hours, its clients and how many people share the key. Behavioural Intelligence, which is coming soon, will learn that behaviour for each account, so a seat that starts serving many people stands out before its buyers complain.
Our token resale guide sets out the signals and the steps to confirm resale without hurting real customers. Our ban evasion guide covers the seller who comes back on a new account.
Book a demo to see how a resale flag reaches your review queue before it reaches a customer.
Every token an AI platform serves runs on compute someone pays for. Token fraud is when the person using it is not the person you sold it to. This guide maps the four ways that happens and links to a guide for each.
GuideA resold account looks like one customer and behaves like a hundred. This guide sets out the signals in your own logs that give it away, how to confirm resale without hurting real customers, and what to do once you find it.
GuideA flat-rate plan priced for one person becomes cheap token supply when it is shared. Anthropic answered in 2025 with a weekly cap on Pro and Max subscribers, and software that shares subscriptions as an API appeared five months later.
CaseA ban removes an account, not the person or the setup behind it. On an AI platform the strongest evidence that they are back is how the new account uses your API.
GuideOpenAI has not published a policy page naming sub2api. Tibo, who works on Codex and ChatGPT at OpenAI, posted on 21 August 2026 about this use. He said turning a subscription into API traffic for many users is not supported and gets flagged by fraud prevention. sub2api's own README warns that it may violate upstream providers' terms of service.
Usage depends on the model, the task and whether it runs locally or in the cloud, so large tasks drain the allowance quickly. In June 2026 OpenAI also reported that its anti-abuse systems had rate limited some accounts by mistake. In August it said many users reporting changed limits were running sub2api, which spreads one subscription across many users.
Yes, according to Tibo's post. Signing in with ChatGPT through the official clients or an open-source client that supports it is fine. What OpenAI objects to is turning the subscription into API traffic that is re-served or shared with other people.
Measure a new rule's false positive rate before it limits anyone, send each flag to a person for review, and tell customers when you limit them. Act on behaviour across several signals, such as hours of use, clients and networks, and never on a complaint about limits alone.
OpenAI's documentation describes Sign in with ChatGPT as subscription access, drawing on the plan's included allowance, to which weekly limits may also apply. An API key is usage-based access, billed at standard API rates.
