Chinese developers can reportedly access Anthropic’s Claude models through a gray market of overseas API proxies, despite the company’s regional restrictions. An analysis by Zilan Qian of the Oxford China Policy Lab describes services known as “transfer stations” that forward requests through servers in supported countries and relay the answers back.
Anthropic uses controls including geoblocking, phone and payment checks, billing-address verification, ownership restrictions, and, for some accounts, identity documents paired with a live selfie. Proxy operators sit between the developer and Anthropic, so the original user may never pass those checks directly.
Some services advertise Claude tokens at about 10% of the official price. Qian’s analysis says operators can reduce costs by exploiting promotional credits and, in some cases, substituting a cheaper model without clearly telling the customer. That means buyers may receive uncertain service while also giving prompts and data to an untrusted intermediary.
The arrangement creates a broader monitoring problem. If many unrelated users appear behind a smaller set of overseas accounts, Anthropic has less reliable information about who is using its models and for what purpose. The report argues that this can weaken both regional access rules and safety systems intended to identify misuse. Blocking individual proxies may not be enough because the supply chain is modular and operators can replace accounts, payment methods, and servers.