Underground Proxies in China Sell Discounted Claude Access, Evading Anthropic's Geofence

A thriving gray market in mainland China is providing cheap, unauthorized access to Anthropic's flagship Claude models, circumventing the artificial intelligence lab's geographic firewalls and billing systems.

According to an investigation by Zilan Qian, a researcher at the Oxford China Policy Lab, mainland developers and consumers are tapping into the proprietary models through intermediary networks termed "transfer stations." These offshore API proxy hubs effectively mask user traffic, bypassing Anthropic's platform blocks against Chinese IP addresses.

Arbitrage and Model Swapping

The proxy services offer access at steep discounts, with token pricing often falling between 70% and 90% below Anthropic's standard catalog rates. Operators maintain these low prices through a mixture of promotional credit harvesting, pooled multi-user account subscriptions, and regional price exploitation. In some instances, vendors reduce their overhead by quietly downgrading user queries to less capable, cheaper models behind the scenes.

The gray-market platforms also remove common technical hurdles for domestic users. Buyers can fund transactions directly via local payment rails such as WeChat Pay and Alipay, eliminating the necessity for international credit cards or dedicated virtual private networks.

Enforcement and Distillation Risks

Beyond simple revenue loss, these relay networks strip frontier AI labs of visibility into who is using their systems and for what purpose. Because queries originate from clean overseas servers, Anthropic's internal abuse-prevention systems struggle to flag terms-of-service violations or trace bad actors.

The dynamic poses a persistent challenge for Western AI companies attempting to comply with export controls and unilateral deployment restrictions. Unmonitored proxy infrastructure can be readily weaponized to distill Claude's outputs into domestic Chinese models, run automated vulnerability research, or facilitate credit card fraud at scale.

As the performance gap between leading American models and accessible alternatives remains a core competitive friction, third-party arbitrage networks are likely to stay a step ahead of routine platform enforcement.