My Claude Account Got Banned
Core Highlights
A developer who has long used Claude published a post on a WeChat public account, documenting how her account got banned from start to finish. The trigger was a SEPA verification flaw in Anthropic's payment system that some users exploited to get service for free, a "zero-yuan purchase." Afterward the company mass-reclaimed those exploited accounts and banned associated ones in a sweep, and the author's main account, used for over half a year, was banned on July 29. The article does not stop at complaining; it offers a pragmatic list of domestic alternatives available today, framed by the realization that the model landscape has shifted underneath everyone who assumed one provider would always be the safe choice.
What Happened
The incident began with a defect in Anthropic's payment path around SEPA (European bank transfer) verification. Some users leveraged it to use the service without paying, creating a "zero-yuan purchase." Once the platform noticed, it began reclaiming those exploited accounts in bulk and went further to ban other accounts linked to them. The author's own account, which was paid normally and used for more than half a year, was also caught in the sweep. For her, the loss was not only the balance but also half a year of accumulated workflows, configurations, and habits that do not transfer cleanly to another tool. The episode shows how a single payment-side bug can cascade into account-level penalties that reach entirely innocent users who merely shared a payment rail or an organization with the offenders.
Alternatives and Recommendations
The author's conclusion is blunt: this is no longer an era where Claude stands alone. For coding she recommends Kimi K3 and GPT-5.6 Sol; for office work she suggests WorkBuddy paired with Kimi K3. Behind this recommendation is an observation about the shifting model landscape: domestic models can now reach the top tier using roughly one-twentieth of the compute, so daily coding and document processing no longer have to depend on overseas closed-source models. The practical takeaway is that capability has decentralized, and users who once felt locked in now have credible, cheaper exits that cover most of what they actually do in a workday.
Comparison with Competitors
Claude still has a reputation for long context and code quality, but price and availability swings are real weaknesses that this incident exposes brutally. Kimi K3 is stable on Chinese-language and long-text tasks, while GPT-5.6 Sol can already go toe-to-toe with flagship models on coding benchmarks. For domestic users, avoiding payment and compliance risks while getting capable performance at lower cost is becoming a realistic option rather than a compromise. The contrast is sharp: one vendor's policy change can erase months of setup, while a diversified toolchain keeps work moving even if a single account is suspended.
Industry Impact and Use Cases
This ban incident is a reminder that betting critical workflows on a single overseas closed-source service carries risk; both payment compliance and account policy can change at any time. Spreading the toolchain across domestic and open-source options is both cost reduction and risk hedging. For creators and developers, building replaceable, portable workflows is more valuable than blindly chasing the newest model. Put simply, there are many capable models, but staking your entire setup on one company's account is a risk this incident has brought into sharp relief. The broader lesson is that resilience now matters as much as raw capability when choosing where to build, and a workflow that can survive a ban is worth more than a slightly higher benchmark score.
For the broader developer community, the post is less a lament and more a field report from someone who hit a wall that anyone could hit. The lesson is not that Claude is bad, but that depending on a single external account for your daily work is a single point of failure you do not control. The author's switch list is also a signal that the domestic model bench has quietly caught up on exactly the tasks people do most: editing code, summarizing documents, and drafting prose. When the gap shrinks to the point where a one-twentieth compute budget reaches the top tier, the economic argument for paying a premium for a single vendor weakens fast. The healthy response is not loyalty to one flag but a portable setup that keeps working when any single provider changes its mind.
The episode also exposes a structural fragility in how overseas AI services are billed in China. A payment rail meant for one region became the lever that swung account standing for users far away, and the cleanup was blunt. For anyone running a business on top of such a service, that is a due-diligence item, not a footnote. The author's pivot is therefore a template: keep a primary tool, but maintain a working alternative you have actually used, so a ban is an inconvenience rather than a stop.