China's AI Secret: Dodging AI Bans for Top Models

Users in China are circumventing strict access controls to utilize AI models like Anthropic's Claude, which they consider the world's most capable AI assistant. Startups and enthusiasts have developed increasingly advanced workarounds to access these tools, despite Anthropic's efforts to block them. This has fueled a black market where accounts and API tokens are bought and sold on platforms like Taobao and Telegram.

Anthropic employs various detection systems, including identity verification, to enforce its policies, leading to account suspensions for users in China. The demand has spurred the emergence of "transfer stations" that act as intermediaries, reselling API access to Chinese users. Programmers, in particular, favor these Western models over domestic alternatives, citing a performance gap.

Recent identity verification measures, requiring government-issued IDs, have shifted the workaround market towards fake identities. Despite these challenges, users continue to find ways to access Claude, demonstrating a persistent demand for advanced AI capabilities in China. However, this reliance on unsanctioned tools exposes users to security risks and raises new AI safety concerns.