A gray market for Anthropic’s Claude models is operating in China despite the company’s access restrictions, according to The Decoder, which cites a detailed analysis by Zilan Qian, a researcher at the Oxford China Policy Lab, published by ChinaTalk. The report says Chinese developers can buy Claude tokens for about 10% of the official price through services known in Chinese developer circles as “transfer stations.” These services act as API proxies: they receive a user’s request, route it through servers outside China as if it came from an approved location, and pass the model’s response back to the user. The Decoder says the setup removes several barriers that would otherwise block direct access. Users do not need a virtual private network or a foreign credit card, and they can pay in Chinese yuan through WeChat or Alipay. The report says some popular transfer stations are tracked in community directories and ranked by price and availability. Anthropic’s controls, as described by The Decoder, include checks on phone numbers, foreign credit cards, and billing addresses. The company also bans companies that are more than 50% owned, directly or indirectly, by entities based in unsupported regions such as China. For select users, the report says Anthropic requires ID verification with a live selfie. Qian’s analysis frames the transfer station as one part of a broader supply chain. Upstream actors reportedly include account brokers that mass-register Anthropic accounts, SMS verification platforms that supply foreign phone numbers, and specialists who study detection methods. Downstream, developers, companies, and resellers market access on Chinese e-commerce platforms including Taobao, according to The Decoder. The user base is not described as limited to major AI labs. Qian says customers likely include Chinese AI labs seeking to distill Western models, along with students, researchers, developers, tech employees, companies, app makers, and hobbyists. The distillation claim is important but should remain hedged: The Decoder reports it as Qian’s assessment of likely usage, not as a confirmed list of buyers. The report also says the market is resilient because many participants control only one or two pieces of the chain. If one provider is banned, Qian argues, the remaining account pools and customer channels can remain in place, allowing replacements to appear within hours. The most sensitive part of the report concerns identity checks. The Decoder says Qian found workarounds for KYC-style verification, including fake IDs generated with AI tools, deepfake techniques aimed at biometric checks, and, in some cases, real people recruited from low-income countries for verification markets. The report notes that those claims are based partly on informal conversations and publicly available sources, so they should be read as evidence of an ecosystem rather than a fully independently verified map of every participant. Who benefits: Chinese developers and resellers benefit from cheaper and easier Claude access, according to the report. Transfer-station operators and adjacent account, phone-verification, and resale services also benefit if the supply chain functions as Qian describes. Who's exposed: Anthropic is exposed because the report suggests its China controls can be routed around by intermediaries. Any AI company relying on similar geoblocking, payments, or identity checks faces the same enforcement question, though this cluster only documents claims about Claude access.