The tug-of-war over account bans between Chinese users and Claude: paying for accounts, getting banned repeatedly, yet still not leaving

Is Anthropic a barbaric company? Ask any Chinese user who has lost an account, and the answer you get will most likely be yes.

By normal business logic, if a company indiscriminately and frequently bans users in a specific region—even banning them right after they have topped up—it would most likely soon be spurned by consumers and face mass cancellations. Yet in Anthropic's world, Chinese users not only have not left; they keep pouring in one after another, buying a new account every ten days to two weeks for a few dozen yuan, adding a subscription of at least the Pro tier to each account, and willingly donating money to the A÷ on the other side of the ocean.

The most classic Claude account-ban meme circulating online

Most of those account-ban moments that catch Claude users off guard have actually been summed up by netizens. Claude's strict risk-control system often strikes domestic users precisely across multiple dimensions.

First is the cleanliness and consistency of the network environment. A proxy node with the slightest flaw or a frequently changing IP address will bring risk to the user; second is anomalous payment behavior. Domestic users cannot directly bind overseas credit cards, so they turn to Taobao top-up agents, virtual credit cards, or SMS verification platforms to register. This kind of abnormal billing association and bulk operation will all be flagged as risky behavior;

Beyond that, a series of factors such as the system time zone, browser language, and login time are said to be among the criteria as well. The official side will not disclose the weight of these factors, but some users, in order to use their accounts safely, have long since started setting their system time to European and American time zones and then following a Claude-specific daily schedule: it is best not to use it during the day, 9 a.m. Beijing time is the most dangerous time, once work hours are over you can log in, and heavy use begins at night and continues into the early morning.

Finally, there is what can only be called the metaphysical "behavior-identity recognition." On social platforms such as Xiaohongshu, Bilibili, and Zhihu, anecdotes about Claude banning accounts because users behave and speak like Chinese people have long since become a spectacle. Some joke that Claude's underlying database has no category for "suspected Chinese" at all, only "confirmed Chinese and Chinese who are relatively good at disguising it." In those famous Claude jokes, typing a Chinese comma in a prompt, or using an em dash with a gap in the middle, or even just telling Claude that you have to work overtime this weekend or that your apartment has shared-area square footage, may accidentally trigger an alarm and bring about the end of your account.

How many heartfelt words are said in a fake email

Anthropic's official report shows that between January and June 2026, a total of 11.4 million accounts worldwide were banned, and of 398,000 appeals, only 42,000 succeeded. It is unclear whether the extremely low appeal rate and success rate prove that these more than ten million accounts truly "deserved to die," or whether they show that even Anthropic's loyal users cannot "divine the imperial will" before the mysterious iron fist falls.

But in the eyes of consumers, this has to some extent become Claude's best reverse advertisement—the fiercer the bans, the smarter the algorithm; and the smarter the algorithm, the more it highlights its absolute dominance in the model ecosystem.

Under Anthropic's relentless domestication, domestic users of Claude today are all like experts in a cat-and-mouse game. Click into an AI community or chat group and ask how to prevent Claude from banning you, and you will have countless options to choose from: residential IPs, exclusive nodes, virtual cards, billing address in Wyoming... If you are not sure whether your account is safe, you can also go to an independent developer's webpage called "Fuck Claude" to test yourself and see whether you are pure-blooded Chinese.

As can be seen, apart from the ban reasons mentioned above, whether you use domestically branded devices, domestic browsers, and Chinese fonts, and even the rendering style of Emoji, are also within the criteria.

There are many other similar Claude ban pre-detection websites

Why, when everyone knows accounts will be banned, do they still keep giving Claude money again and again? Ciwei Gongshe (ID: ciweigongshe) asked several heavy AI users in various industries, and their answers differed, but all were rather bitter:

As someone who has frequently experienced bans firsthand, international student Sonia has lost count of how many times she has been banned. "With the first account I paid no attention at all to whether the IP was clean, and it was banned very quickly; the second account was even more absurd—I was in Hong Kong at the time, using an overseas node, and when I logged in it prompted me for phone verification, and right after I scanned the code the account was gone."

Most ban memes are not real chat records, but these images convey the same message: make one mistake before Lord Claude, and you may suffer undeserved disaster.

After getting a third account, Sonia began to learn her lesson and looked up a large number of guides before using it: use incognito browsing, strictly switch the computer and phone to the same node, and carefully log out of the webpage before disconnecting the node each time. Yet there was no guarding against it—while running a project at noon she took a nap, and the computer screen turned off, causing the node to disconnect automatically. Sonia guesses that perhaps because a change in the network environment was detected, by the time she woke up the account had already been shown as banned.

By the fourth account, Anthropic showed its thuggish true colors even more: it banned Sonia within a minute after she topped up her balance. Sonia's case shows that your loyalty to Claude cannot buy back Claude's kindness toward you—"the first sword upon coming ashore is to strike down the one you love." According to the IPO prospectus Anthropic submitted to the U.S. Securities and Exchange Commission, in the first nine months of 2026, about 5% to 7% of Anthropic's revenue came from accounts later determined to be in violation.

Overseas, there are also websites collecting cases of users whose Claude accounts were banned. Image source: bannedbyanthropic.com

Most of the banned users registered on that site are developers. Besides Chinese, there are also Americans, Singaporeans, Kuwaitis, and others.

Another user, Xiao W, is a designer, and Claude is a daily production tool for him. He was first banned at the end of 2025, after using the account for about four months; one day when he logged in, he was suddenly prompted that the account was restricted. During that period he had not changed nodes, had not changed his payment method, and did not feel he had said anything out of line. Xiao W said the only change was that he took on a rush job that week and logged in and used it more frequently than usual. "Maybe the system thought I didn't look like a normal user, but I was just rushing work."

When he was first banned, Xiao W seriously wrote an appeal letter, but it did not succeed; he only received a template reply. After that, Xiao W never considered appealing again—in his eyes, now if an account lasts half a month, that counts as breaking even, and if it lasts a month, that counts as burning incense in gratitude. In my view, Xiao W's situation on Claude is about the same as that of an overseas undocumented worker who has no way to defend his labor rights.

Now Xiao W uses Claude very practically: he only tops up monthly, exports important materials as he goes, and is always on guard against Anthropic pulling the rug out from under him. But even after experiencing many bans, Xiao W still has not migrated his main work to other AIs. For him, the workflows and work habits he has accumulated on Claude over the long term, as well as the level of productivity he can achieve with Claude, are beyond what other large models can reach.

An image attached by a netizen to a post about his own Claude account being banned. The first impression Claude's work leaves on many users is simplicity and elegance. Image source: Xiaohongshu @James

Another creative worker, Zoe, said that after returning to China from New Zealand, she gradually stopped using Claude. In New Zealand, her use cases for Claude were very clear: when cross-cultural communication required clarifying the background of a matter, or when she wanted only a definite answer to a question and did not need AI to help her brainstorm. Claude's restraint, stability, and refusal to casually add drama were great advantages.

After returning to work in China, Zoe uses ChatGPT more to help her brainstorm and to share some personal life reflections. Zoe told me that her best New Zealand friend, by contrast, has always relied heavily on Claude: a typical STEM student who strictly follows the life logic that A is A and B is B, and who does not seem to need much emotional or feeling-related support.

From Anthropic's attitude toward domestic users, Claude's product style, and its user profile, we can roughly see the underlying logic of this "tormented romance": the relationship between Anthropic and users is not like a traditional service relationship, but more like a screening. Anthropic is like an arrogant luxury goods store that takes pleasure in humiliating customers, with a clerk in a suit standing behind the glass, looking at everyone as if they were there just to enjoy the air conditioning.

The real NPD is best at pretending to be innocent when questioned

But even after repeated setbacks, domestic professional users still cannot completely let go of Claude. From user reviews on social media, it can be seen that after Claude 3.5 Sonnet broke into the mainstream, its subsequent versions established a highly recognizable moat in logical reasoning, complex code architecture generation, and long-text comprehension. The irreplaceability of a productivity tool gives Anthropic extremely high pricing power, and also forms the confidence that allows it to treat users brutally. On the geek high ground of commercial competition, when the technological gap is significant enough, user experience is squeezed into a secondary position.

I love you, but I should go.

Behind the "Claude ban jokes" full of punchlines and bitterness are the real needs and helpless self-mockery of China's heavy AI user community. One netizen compared Claude to "a tsundere but sensible genius young lady," and also said that between users and Claude stands a "perverted and unreasonable stepfather," namely Anthropic.

Netizens' anime-style depiction of Claude

Of course, even overseas, Anthropic is a controversial, mixed-reputation alternative Silicon Valley figure. This company always stirs the nerves of the entire tech world with an almost obsessive, even Quixotic "cult" posture. They hold high the banner of Constitutional AI and style themselves as the ultimate whistleblower on risks to human survival. Whenever they release a new model, they always put on an attitude of treading on thin ice and trembling with fear, claiming that because they worry the technology is too powerful and may be misused, they have to conduct lengthy and rigorous safety alignment.

On European and American tech forums, satirical jokes and "ban literature" about Anthropic are no less abundant. Independent developers in Silicon Valley are likewise bitterly criticizing Anthropic's black-box review mechanism and high API pricing. Someone joked: "Anthropic's safety vision is: before ensuring humanity is not destroyed by superintelligent AI, first make all legitimate developers go bankrupt because they cannot pay the bills or are banned for no reason."

Many netizens think Anthropic's so-called safety concerns are just hypocrisy with ulterior motives

Reality is often full of black humor—Anthropic, which always puts on an ivory-tower academic and saintly face, now also wants to tear off its aloof persona and race toward Nasdaq, the ultimate judgment seat of the capital market. And in the 335-page prospectus, about 100 pages warn of the potential risks its technology may pose to humanity and society.

It is unclear what place excluding the Chinese people, who have always loved peace and pursued universal harmony, occupies in Anthropic's "risk-focused" narrative.

Sonia, Xiao W, and Zoe in this article are pseudonyms

This article comes from the WeChat public account "Ciwigongshe" (ID: ciweigongshe), author: Maojin, editor: Yuanzhang, published by 36Kr with authorization.

In European and American tech forums, satirical jokes and "ban literature" about Anthropic are no less abundant. Independent developers in Silicon Valley are likewise bitterly criticizing Anthropic's black-box review mechanism and high API pricing. Someone joked: "Anthropic's safety vision is: before ensuring humanity is not destroyed by superintelligent AI, first make all legitimate developers go bankrupt because they cannot pay the bills or are banned for no reason."

Many netizens think Anthropic's so-called safety concerns are just hypocrisy with ulterior motives

Reality is often full of black humor—Anthropic, which always puts on an ivory-tower academic and saintly face, now also wants to tear off its aloof persona and race toward Nasdaq, the ultimate judgment seat of the capital market. And in the 335-page prospectus, about 100 pages warn of the potential risks its technology may pose to humanity and society.

It is unclear what place excluding the Chinese people, who have always loved peace and pursued universal harmony, occupies in Anthropic's "risk-focused" narrative.

Sonia, Xiao W, and Zoe in this article are pseudonyms

This article comes from the WeChat public account "Ciwigongshe" (ID: ciweigongshe), author: Maojin, editor: Yuanzhang, published by 36Kr with authorization.