In China, access to Claude was sold at a 90% discount through gray-market schemes

In China, the grey market for proxy servers was trading access to Claude — Anthropic's flagship model — at a 90% discount from the official price. This was revealed by research conducted by Ziyu Qian, a researcher at Oxford China Policy Lab.

How the scheme works

Traders were using multiple profit-making techniques simultaneously. On stolen Anthropic accounts, they launched proxy servers, giving users access to Claude at a 90% discount. But that was just the beginning — next, they replaced Claude with cheaper models (GPT-3.5 or local ones) and pocketed the difference. And most importantly: user queries and responses were resold to third parties for distilling their own AI models.

• Using stolen Anthropic accounts
• Replacing Claude with cheaper models
• Collecting and reselling user data
• Official price minus 90% discount = profit on every customer

Why this is a problem

For users, it means their queries fall into uncontrolled hands. Their data is used to train competing models. For Anthropic — it's a loss of control over the model in one of the world's largest AI markets and gross losses. The research showed that the scheme was active and finding demand — users liked the cheap price, and sellers liked the profit. Chinese Internet law enforcement typically doesn't aggressively pursue such schemes if they target foreign companies.

What it means

This demonstrates a classic conflict between market openness and IP control. Anthropic cannot effectively protect Claude in places where the state is not interested in protecting it. For Chinese companies, mining data for their own models is profitable. The global AI economy is still a wild west without unified rules.