A black market for Anthropic's Claude AI model, blocked in China, has grown to include thousands of token resellers. These resellers are fueling illegal AI training and distillation attacks, according to The Information.
TL;DR
- A black market for Claude access has emerged in China, with resellers offering access at less than half the official price.
- Illegal accounts are used for distillation training, allowing Chinese companies to train their own AI models using Claude's responses.
- Anthropic's restrictions have led to sophisticated circumvention methods, including the use of paper companies and resold tokens.
What happened
Anthropic's Claude AI model, officially blocked in China, is being resold on a black market by token resellers. These resellers establish servers in countries where Claude is permitted and gather numerous accounts to sell access to Chinese users at less than half the official price, according to The Information.
The black market for Claude access is estimated to include thousands to tens of thousands of resellers. These resellers use methods such as creating AI brokerage firms, using bulk email addresses, and paying with stolen credit cards or stablecoins to secure accounts.
Illegally obtained accounts are used for distillation training by companies like Moonshot AI and DeepSeek. This process involves flooding Claude with questions to quickly train their own AI models using the responses.
As Anthropic tightens restrictions, circumvention methods have grown more sophisticated. Chinese tech firms establish paper companies in Southeast Asia, the Middle East, and the U.S. to sign enterprise agreements with Anthropic or Amazon Web Services. Leftover tokens from overseas startups are also resold to China.
Why it matters
For AI/ML developers, the black market for Claude access highlights the demand for advanced AI models in China. It also raises concerns about the ethical implications of using illegally obtained accounts for AI training.
For startup founders, the growth of the black market indicates a potential market for AI models in China. However, it also highlights the challenges of operating in a regulated environment and the need for robust security measures to prevent unauthorized access.
For tech investors, the black market for Claude access underscores the potential for AI models in China. However, it also highlights the risks of investing in companies operating in a regulated environment and the need for due diligence to ensure compliance with local laws and regulations.
Key facts
- Anthropic's Claude AI model is officially blocked in China and never launched services there.
- Anthropic suspended services for overseas organizations with over 50% direct or indirect ownership by entities from hostile countries, including China, in September last year.
- Six out of 30 companies in an office building in Haidian District, Beijing, are engaged in reselling AI access rights, according to The Information.
- Token resellers within China are estimated to number in the thousands to tens of thousands.
- Token resellers offer Claude access APIs to Chinese users at less than half the official price.
- Illegally obtained accounts are used for distillation training by companies like Moonshot AI and DeepSeek.
- Chinese tech firms establish paper companies in Southeast Asia, the Middle East, and the U.S. to sign enterprise agreements with Anthropic or Amazon Web Services.
- Leftover tokens from overseas startups are resold to China.
Context
Anthropic is a leading AI company based in the U.S., known for its advanced AI models like Claude. The company has faced challenges in operating in China due to regulatory restrictions and concerns about national security.
The black market for Claude access highlights the demand for advanced AI models in China, despite regulatory restrictions. It also raises concerns about the ethical implications of using illegally obtained accounts for AI training.
The growth of the black market for Claude access underscores the potential for AI models in China. However, it also highlights the challenges of operating in a regulated environment and the need for robust security measures to prevent unauthorized access.