Anthropic cracks down on Chinese workarounds to Claude

Anthropic is tightening its enforcement against unauthorized access to its AI services from China, targeting methods used by Chinese companies — including Ant Group Alibaba Group Holding Limited ↗ 1.28% — to circumvent its restrictions through overseas entities and cloud infrastructure, the Financial Times reported on Wednesday.ft+1

The AI company, which builds the Claude family of models, has identified a range of workarounds that Chinese firms have employed to gain access to tools like Claude Code despite being barred from its services. These include routing usage through overseas subsidiaries and third-party cloud providers, according to the report.ilsole24ore+1

Tracking Proxies and "Transfer Stations"

To detect unauthorized usage, Anthropic has been monitoring accounts for telltale indicators such as computer time zones set to Chinese regions. Earlier this year, a Reddit user discovered that Claude Code had embedded hidden markers in its system prompt that could flag users routing through Chinese proxies, checking for time zones like Asia/Shanghai and whether proxy URLs were linked to Chinese AI labs.reddit+1

Anthropic's Thariq Shihipar described the mechanism as "an experiment we launched in March that was meant to prevent account abuse from unauthorized resellers and protect against distillation". The company said it has since rolled back that particular tracking method while deploying stronger protections.aiweekly+1

Anthropic is also cracking down on so-called "transfer stations" — proxy services that acquire Claude API tokens abroad and resell access to users inside mainland China, often at steep discounts. These operations have created a thriving underground economy, with Claude access sold on platforms including Taobao and Telegram channels.wired+1

Escalating Tensions Over Chinese Access

The latest enforcement push comes amid a months-long series of confrontations between Anthropic and Chinese AI interests. In February, the company accused DeepSeek, Moonshot AI, and MiniMax of using roughly 24,000 fraudulent accounts to interact with Claude more than 16 million times in an effort to train their own models. In June, Anthropic alleged that Alibaba had conducted the "largest initiative to illegally tap into Claude's functionalities," using 25,000 fake accounts to generate over 28.8 million interactions.wsj+3

The ripple effects have extended beyond Chinese AI labs. In June, JPMorgan Chase ↘ 0.84% removed Claude models from its list of approved tools for Hong Kong staff after reviewing Anthropic's restrictive usage terms.reuters

Standing Alone Among AI Firms

Anthropic has said it is the only frontier AI company that restricts sales to companies controlled by the People's Republic of China, including subsidiaries incorporated outside the country. The company first announced a ban on majority Chinese-owned entities in September 2025, extending restrictions to cover companies more than 50 percent owned by entities headquartered in unsupported regions.scmp+2

The enforcement challenge underscores the difficulty of policing access to cloud-based AI services across borders, even as geopolitical pressure mounts on American AI companies to prevent their technology from reaching Chinese competitors.