While Anthropic tightens its restrictions and introduces identity verification, intermediary services, black markets for accounts, and a whole industry dedicated to bypassing these barriers have emerged in China.
Are you choosing a modular home? Fixed prices are listed on the website.
Take a free tour of 15 completed homes! Offer valid until July 10th.
Despite efforts by the American company Anthropic to limit access for users from China to its AI assistant Claude, a vast ecosystem of workarounds has developed in the country. This ranges from selling pre-made accounts and fabricated identities to specialized intermediary services that provide stable access to the language model. According to an investigative report by Wired, demand for Claude in China remains so high that a shadow market has essentially formed around it.
Anthropic, founded by former OpenAI employees and known for its strict safety policies, does not officially provide access to Claude for users in China. Unlike many other Western services, the company goes beyond simple geo-blocking and IP address checks. It actively identifies suspicious accounts, blocks users who employ VPNs and proxies, and has recently begun requiring identity verification through the Persona service using government-issued documents.
However, Chinese users continue to find ways around these restrictions. The simplest approach involves using VPNs, foreign phone numbers, and international payment tools. More complex options include purchasing pre-registered accounts on Chinese marketplaces like Taobao and Xianyu, as well as through specialized Telegram channels. According to cybersecurity experts, a thriving underground market for Claude Pro and Claude Max subscriptions has formed.
So-called “relay stations” have gained particular popularity. These services are hosted in countries where Claude is officially available, acquire corporate access to Anthropic’s API, and then resell it to users inside China. From the client’s perspective, it appears as a regular chat interface: a query is sent to a local website, forwarded through an intermediate server to Claude, and then the response is sent back. For professional users, especially developers, this approach has proven more convenient and even cheaper than official access.
Demand for such services surged following the emergence of a new generation of AI agents, such as OpenClaw. These systems handle more complex tasks and require significantly more computational resources and queries to language models. For programmers who actively use Claude Code and OpenAI Codex, stable access to cutting-edge models has become a critical tool for their work.
At the same time, many Chinese professionals consciously prefer American models over domestic alternatives. Employees of tech companies and research institutions interviewed by Wired stated that they use Claude for programming far more often than Chinese models like DeepSeek or Z.ai. According to Zilan Qian, a research fellow at the Oxford China Policy Lab, Chinese developers estimate that domestic language models lag behind American ones by about 6 to 9 months, particularly in software development tasks.
Amid the technological rivalry between the US and China, Anthropic’s leadership has repeatedly described Chinese companies’ access to advanced AI models as a national security threat. The company’s CEO, Dario Amodei, previously argued that Chinese firms could use Claude’s responses to train their own models through distillation—a process of transferring knowledge from one neural network to another. Last week, Anthropic publicly accused Chinese tech giant Alibaba of engaging in such practices.
A paradoxical consequence of these restrictions is that they have only spurred the development of new evasion techniques. After the introduction of mandatory identity verification through the Persona service, the market shifted from selling accounts to selling pre-verified digital identities. According to cybercrime investigators, Chinese-language Telegram channels are now actively advertising Claude accounts that have already passed the document verification process.
The scale of these workarounds has become so significant that it may have skewed global statistics for Claude’s adoption. For example, Singapore, with a population of about 6 million, regularly ranks among the world leaders in per capita usage of the service. Experts believe that a substantial portion of this traffic may actually come from users in China who are leveraging Singaporean servers and intermediary platforms.
About the Author
