AI API Resellers That Vanish With Your Balance: Warning Signs, Exposure Math and a Safer Pattern (2026)
How AI API reseller shutdowns happen, the signals that come first, how to calculate your real exposure, why funds are rarely recoverable, and a small-top-up routine that caps the loss.
1. How reseller shutdowns actually unfold
The pattern repeats. A reseller appears with prices far below the maker's list, builds a following, and then pushes volume hard with prepaid bundles and bonus credit. Balances accumulate. Then one of several things happens: the upstream account is cut off, the operator decides to stop, or the whole thing was always designed to collect and leave. Support replies slow down, the site goes offline, the group is dissolved, and balances go to zero. Some shutdowns are not malicious, just fragile businesses built on accounts that were never meant to be resold. For the customer the result is the same. This is why the useful question is not whether a particular reseller is honest, which you often cannot know, but how much of your money would be exposed if it stopped tomorrow.
2. The warning signs that usually come first
None of these proves anything alone, but two or three together should stop you from adding money. No identifiable company, just a group and a payment code. Private transfer or collection in a chat as the only way to pay. Constant offers of the 'top up X, get Y free' type that reward a large prepayment. A price so low that no plausible cost structure supports it; sustained loss-making prices are often the volume push before an exit. Reluctance to be tested, or billing you cannot see, such as a quota that drops with no per-call record. Support that works only through one person's personal account. Finally, a long history is weak reassurance, because long-running operations have also disappeared.
3. Calculate your real exposure
Exposure is the balance you hold, not the discount you were promised. A simple way to think about it: if your average use is a certain amount per week and you prepay a quarter, you are holding about twelve weeks of risk to save a small percentage. Work out your weekly spend from your own logs, then decide how many weeks of balance you are willing to lose without it hurting. For many people that is one to two weeks. Compare that to what a bonus promises. A twenty percent bonus on a prepayment you would not finish for three months is not twenty percent, because you are lending the money for three months with no collateral and no guarantee of return. If the arithmetic only works when you prepay big, the discount is the price of the risk.
4. Why recovery is unlikely, and what that means
A shutdown usually leaves nobody to pursue. With no company, no verified identity and only anonymous collection, there is no one to write to and no process to follow. Private transfers and crypto payments cannot be reversed. Even card payments have limits: disputes are possible on some channels, but the time window is limited and success depends on the payment provider's rules and on what you paid for, so do not treat a dispute as a plan. The consequence is that the useful defence is entirely before the payment: keep the balance small, choose a channel that leaves a record, and decide in advance that any balance above your comfort level stays in your own account rather than the reseller's wallet.
5. A top-up routine that caps the loss
Use a routine instead of a judgement. First, test with a small amount, then run real work on it, checking latency, billing per call and model behaviour. Second, set a weekly or fortnightly top-up equal to expected use, not more. Third, record the balance after each top-up and each week, so a sudden unexplained drop is visible. Fourth, keep a second provider configured and tested, so a shutdown is an outage of an hour instead of a week; with OpenAI-compatible endpoints this means holding two base URLs and two keys in configuration. Fifth, keep keys for each project separate so you can revoke one quickly. None of this requires distrusting your provider; it simply ensures that if the provider fails, the cost is a small and known number.
6. A due-diligence checklist you can run in ten minutes
Before the first top-up, check five things. Is there a company name and public information you can verify independently? Does the payment channel leave a record? Is billing per token with a visible ledger? Can you check the model with a third-party fingerprint tool such as probe.cocodot.co, which works against any endpoint? Are fees and limits stated in writing, including what happens to a remaining balance? Then run a small real workload and compare the deducted amount with the listed price over ten calls. If anything does not reconcile, stop at that amount. A provider that makes this easy has told you something useful about how it will behave later.
7. What a safer provider looks like
A reasonable provider makes its own accountability easy to verify. It names the company behind the service and publishes how to reach it. It accepts payment through channels that leave a trail. It prices per token and shows each call in a ledger. It does not need you to prepay big to make the economics work. It lets third parties test its models. It states what it will not do, such as invoices, and does not hide limits. cocodot aims to meet these points: company details are on its trust page, top-ups are in small amounts by Alipay or WeChat, billing is per token with a per-call ledger, and a trial credit lets you test before paying. That is not a guarantee that nothing will ever go wrong, which no vendor can offer. It is a structure in which your loss stays small.
8. If a reseller you use stops responding
Act on the assumption that the balance is gone and protect the rest. Move your workloads to your backup provider immediately, since an outage costs more each day. Rotate or delete any keys that were ever used with the reseller, and any credentials you pasted in prompts. Review the payment record and, if you paid through a channel that allows disputes, open one promptly with a clear timeline. Keep screenshots of balances, messages and receipts. Warn others you know who use the same service, without making accusations you cannot support. Then change the rule for next time: smaller balances, more frequent top-ups, and a second provider kept ready.
Warning signs versus reassuring signs
| Area | Higher risk | More reassuring |
|---|---|---|
| Identity | Only a chat group and a payment code | A named company and a public trust or about page |
| Payment | Private transfers or collection inside a group | A channel that leaves a record and can be disputed |
| Pricing pressure | Prepay big, get bonus credit | Pay for what you use, top up when needed |
| Price level | Far below any plausible cost | Close to the maker's list price |
| Verification | Avoids tests; opaque quotas | Per-token ledger; welcomes third-party checks |