Anthropic’s clampdown on access from China via VPN

Anthropic strengthens measures to block access to its artificial intelligence models from Chinaafter several local businesses circumvented the restrictions via VPNforeign clouds and offshore subsidiaries. The reports it Financial Timesciting internal sources.

The US group, which applies among the most restrictive policies in the sector on use in China, has intensified its systems monitoring and verification to identify accounts that violate the terms of service.

Avoidance practices

The practices detected include the use of entities based in Singapore to access APIs and the use of virtual private networks by engineers in mainland China.

According to sources, groups like Ant Financial they would have made Claude business accounts accessible via foreign infrastructures, while ByteDance would have allowed employees to refund individual subscriptions to the platform. These are not explicit violations of the rules, but are practices that conflict with the contractual terms imposed by Anthropic.

Research and development

Access to models, including advanced coding tools, is of particular interest for research and development activities, including “distillation“of models, a technique that allows you to transfer capabilities from more advanced AI systems to smaller, faster models.

Anthropic is also working to limit so-called “transfer station“, which route requests from China to accounts registered abroad, while recognizing the difficulty of controlling all methods of circumvention.