Right now, when you try to sign up for a new OpenAI Codex account, you are no longer greeted by that familiar email verification interface, but by a cold, hard wall – overseas phone number verification is now mandatory.
This is not just a UI interaction change; this is an iron door, and one that has been completely welded shut. Once upon a time, all kinds of "freeloading" secret manuals circulated through tech circles: account generators, account pools, seamless switching. That was the idyllic era of the Token economy, when developers were self-sufficient, sharing the thrill of bypassing restrictions on GitHub.
But now, the door is closed.
Why was it closed? Because OpenAI got sick of being leeched. This is no longer the "jungling" behavior of a handful of tech geeks; it has become an organized, large-scale industrial plunder. When tens of thousands of API Keys roll off the black-market assembly line like snowflakes, and when cloud computing resources are used like a free cash machine running day and night, the "Tragedy of the Commons" inevitably strikes. Resources are limited, but greed is infinite. When everyone pursues their own maximum benefit, the result is the pasture degrades, the gates are locked, and everyone is left outside.
I. Anatomy of an Industry: The Quadruple "Russian Doll" of the Token Economy
Behind this welded-shut door lies an already highly stratified and meticulously operated gray industry. It is like a Russian nesting doll; peel back one layer to reveal another, each layer feeding on the corpse of the one above it.
L1 – The Bottom Layer (Black Zone): Digital Ghosts That Make Something From Nothing
At the very bottom of the chain lies pure criminal activity. The main players here are credit card fraud and batch automation scripts. Black-market actors use credit card information stolen from all over the world to fraudulently purchase cloud services or directly register for API access. For them, this is the true "something for nothing" business. Card issuing, account registration, and token consumption are all done in one seamless motion. They are the fuel suppliers of the entire arbitrage game, and also the segment bearing the highest risk.
L2 – The Technical Layer (Gray Zone): A Frenzied Reverse-Engineering Race on the Web
This layer is closer to home, inhabited by many self-proclaimed "tech geeks" and developers. They don't directly call the paid API; instead, they conduct "packet sniffing" on the ChatGPT web interface through reverse engineering. They analyze request headers, simulate Session sessions, and forge browser fingerprints to forcibly convert the free or subscription-based chat capabilities of the web version into a programmable "Web2API." It sounds cool, but in essence, it's an act of bypassing the billing system and conducting a large-scale "account sharing" fraud.
L3 – The Circulation Layer (Arbitrage Traders): Exchange Rate Gaps and Information Asymmetry
This is the "smartest" layer. They do not produce illegal accounts; they are movers of profit. They exploit pricing differences between countries or buy accounts directly from upstream sources, acquiring genuine subscriptions or API quotas at extremely low discount prices in bulk, then break them apart and sell them retail. Some go even further, exploiting AI enterprise discount agreements or even educational offers obtained by some startups, fraudulently registering massive numbers of sub-accounts to resell. They are parasites living in the cracks of a global pricing system.
L4 – The Top Layer (White Zone): The Difficult Survival of Compliant Aggregation
At the very top are aggregation platforms aiming for legality and compliance. They try to solve the demand problem with business logic – direct official connections, invoices, SLA guarantees. They have the highest costs and thinnest margins, walking on the edge of the tech giants' ecosystems, struggling to act as a "compliant router" between developers and models. They are currently the most stable key, but one that could be snapped at any moment by an upstream policy change.
Of course, reality is far more chaotic than this layered breakdown. The vast majority of relay stations are actually a four-tier hybrid abomination. A little black-card volume pumped in at the bottom, a Web2API setup running in the middle to keep things afloat, and some cheap genuine accounts from Turkey to prop up the facade—if risk controls tighten, they scramble at the last minute to sign an enterprise discount agreement. The proportion of each layer is like a hotpot base recipe: outsiders will never know it, and the owners themselves will never volunteer it.
II. Infighting: When the "First Traitor" Attacks with Full Fury
This seemingly stable pyramid is full of internal mistrust. And the "Killeryou" incident was the knife that tore through this veil.
The sequence of events was simple: "Killeryou," a big player running multiple relay stations in the gray-market world, suddenly launched a full-frontal attack on OpenAI's official community. He openly posted, exposing in detail the cheating methods of his other gray-market peers, including how to batch-register accounts using forged identities and how to evade risk controls, in a tone fierce enough to sound like a righteous whistleblower.
However, he was not a righteous man; he was a player whose interests had been damaged. The truth soon came out: a mass rug-pull on the "black cards" supplied to him upstream led to the banning of three stores under his name, costing him a direct loss of $25,000. With his upstream supplier fleeing and his downstream customers coming for him, he chose to flip the table, cornered and desperate.
This is a textbook case of "thieves falling out." It reveals a brutal truth: when the stability of the gray market completely collapses, and the fraudulent chain it depends on breaks, infighting is the only outcome. There are no rules here, no contracts, only interests. When interests are harmed, the cost of betrayal approaches zero.
III. The Truth About Relays: No Contracts, Only the "Right of Interpretation"
In the black and gray industries, especially those on the internet, no one talks to you about contracts.
1. Top-up Equals "Donation": A Digital Illusion in Lawless Territory
The logic of a relay station is absurdly simple: you send money, which goes through several layers of untraceable money-laundering links. The boss receives the money, then adds a few digits to your ID in the backend database.
This is a game of "Bro, trust me." The Chinese yuan you top up shows on the web panel as a "balance," but that is essentially just a string of code that can be wiped out with one click, or evaporated completely by the "accidental" server shutdown. In the gray-market world, there is no such thing as a "User Agreement." The moment you press the top-up button, you are essentially making a directed donation in a lawless territory. If the boss decides tomorrow to switch to selling pork ribs somewhere else, or if the server gets taken down by the authorities, your balance won't even count as a worthless piece of scrap paper.
2. Multiplier: The Arbitrary Baton in the Boss's Hand
The so-called "multiplier" is essentially the "shadow tax" that relay stations use to harvest dynamically. If the multiplier is 2, then the purchasing power of your 1 Chinese yuan shrinks instantly to 50 cents.
Everyone knows the multiplier can change, but don't expect any "agreement protection." This thing is a dynamic sickle the boss uses to balance costs and harvest profit.
No Law, Only the Backend: The gray market is gray precisely because it is not subject to any regulation. If the boss feels costs are high today, or if upstream black cards get banned, he can just move his finger in the backend, adjust the multiplier from 1x to 10x, and your balance will collapse ten times faster in an instant.
The Logic of Violent Cost-Shifting: Exchange rates fluctuating? The boss wants a new computer? He doesn't need to issue an announcement, let alone ask for your consent. This absolute centralization of arbitrariness is the biggest pitfall of relay stations—your usage cost depends entirely on the boss's mood and greed that day.
3. The Ultimate Price: You're Not Just Paying, You're "Feeding Data"
This is the most insidious and real price in the entire chain. You think you've spent money to buy a service, but you are actually a self-funded miner, personally feeding your most core data assets into the relay station's hands.
Prompt as an Asset: Every single API request is as transparent as the emperor's new clothes in a relay station's backend. Your painstakingly optimized Prompt templates, your company's core business logic, even the code for your closed-source project—all of it lies "awaiting review" in someone else's database.
Self-Funded "Distillation Corpus": The boss, while taking your money, can conveniently export these high-quality conversation datasets. What can this data be used for? It's the most coveted "distillation corpus" for model training. While charging you a fee, he casually places you on the chopping block, turning you into free fertilizer for someone else's model evolution.
IV. The Endgame of Risk Control: KYC
This cold wave is not unique to OpenAI. The recent risk-control actions of its biggest competitor, Anthropic, foreshadow an even harsher future—mandatory real-name verification.
This is not a simple wave of bans, but a combination punch: account bans, a tightening of refund policies, and strong real-name authentication (KYC). Now, to register a stable Claude API account, you may need to submit a government ID (identity card/passport) and cooperate with a real-time selfie verification. This directly blocks countless domestic developers.
This has created a massive structural dilemma. The high wall of network latency, the barrier of international credit cards, and strict real-name requirements—these three mountains are the soil that has nurtured the living space of relay stations. But this parasitic service has never been stable—developers are sinking into a deep anxiety of "interrupted supply at any moment": the service you built based on a certain Key might wake you up to a screen filled only with 429 errors.
The so-called "right to access top-tier foreign models" is being stripped away layer by layer—it is not a privilege, but a temporary pass that can be revoked at any moment.
V. Epilogue: Those 29 "Crying Faces"
At the end of the story, let's turn our eyes to that relay station community channel that collapsed in an instant. Beneath the service shutdown announcement, 29 crying-face emojis lined up in a silent queue.
They are students, indie developers, and micro-entrepreneurs. They just wanted to test a demo at a lower cost, or build a niche AI application. They are not the initiators of this industry chain, but they are the most direct bearers of the cost every time the "Tragedy of the Commons" erupts. Supporting characters in the tide, the ones breathing their last bubbles as they drown.
But this time, their crying faces should not just mourn the death of a relay station. This screenshot should become a piece of evidence—evidence of the predicament we were in, parasitizing our creativity on someone else's ecosystem.
As OpenAI and Anthropic weld one door after another shut, we are finally pushed to the real question: Do we keep looking for the next crack in the wall? Or do we stop and forge the key ourselves?
This question brooks no further delay.
Top comments (0)