Gray Market in China Offers Claude at One-Tenth of Regular Price

A gray market bypassing Anthropic's access restrictions targeting China is expanding. Chinese developers can purchase Claude API tokens at approximately one-tenth of regular prices through intermediary services called 'relay stations.'

Anthropic has historically implemented multiple layers of access controls, including geoblocking that prevents access from China and selfie authentication for identity verification. These measures are positioned as responses to U.S. export controls and efforts to maintain the company's emphasis on safety policies. As artificial intelligence rapidly spreads, access management for advanced AI models has become a matter of concern for countries from a technology security perspective.

Circumventing these restrictions is a network of intermediaries known as 'relay stations.' These operators obtain API access from regions without restrictions and resell it to developers within China. The pricing at approximately one-tenth of regular rates is seen as lowering barriers to entry and accelerating the influx into the gray market.

Analyst Zilan Qian warns that this situation risks undermining both export controls and Anthropic's independently established safety management framework. Anthropic's safety system is designed on the premise of user attributes and usage patterns; as unauthorized access through intermediaries increases, its verification functions become ineffective. This perspective is worth noting.

The importance of this issue extends beyond merely affecting Anthropic's revenue or license management. Access through relay stations makes it difficult to determine who is using the model and for what purpose. From an AI safety perspective, there is a risk that the fundamental design philosophy of confirming developer and user attributes before granting access becomes merely a formality.

The focus going forward shifts to how AI model providers will address this circumvention infrastructure. Since technical restrictions alone can be bypassed, more multilayered measures and policy-level coordination are increasingly required. The existence of this gray market also raises important questions about the effectiveness of AI governance.