Key Takeaways
Anthropic is shutting down the workarounds Chinese companies use to reach Claude, including corporate accounts tied to Singapore subsidiaries, VPN-reimbursed personal subscriptions, and relay services, according to a Financial Times report July 3.
The practices break Anthropic's terms of service rather than US or Chinese law, so enforcement falls to the company itself, which now monitors signals like computer time zones and usage patterns.
CEO Dario Amodei said in February that Anthropic had already walked away from several hundred million dollars in revenue by cutting off firms linked to the Chinese Communist Party.
Anthropic is going after the gray zone. The Financial Times reported July 3 that the company has stepped up efforts to detect and block unauthorized Chinese access to Claude after identifying workarounds that violate its terms of service without breaking any law. Ant Group reportedly gave staff corporate Claude accounts tied to a Singapore subsidiary, ByteDance reimbursed engineers for personal subscriptions accessed over VPNs, and other firms reached Claude through foreign-incorporated units running on cloud infrastructure like Microsoft Azure.
Enforcement without a statute
Because none of this is illegal, the policing is corporate. Per the BanklessTimes account of the FT report, Anthropic now monitors account signals such as computer time zones and usage patterns, and is targeting "transfer station" services that relay requests through overseas Claude accounts. The company's terms have barred any firm more than 50 percent owned by entities in China, Russia, Iran, or North Korea since 2025. Anthropic has framed the systematic querying behind some of this access as distillation, extracting a frontier model's capabilities into competitor training data.
The cost of the line
The company has been explicit that the policy is expensive. Amodei said in February that Anthropic had "forgone several hundred million dollars in revenue" by cutting off CCP-linked firms. The stakes sharpened this week when Meituan open-sourced a 1.6 trillion parameter model trained entirely on domestic Chinese chips, evidence the capability gap the rules protect is narrowing. The move extends a year in which the company's lines kept making news, from the 19-day government-ordered Fable 5 shutdown to the court fight over the Pentagon's ban. The pattern here is hard to ignore: access to frontier AI is being decided as much by terms of service as by treaties.
People Also Ask
How were Chinese companies accessing Claude?
Through corporate accounts tied to Singapore subsidiaries, personal subscriptions reimbursed by employers and accessed over VPNs, foreign units on cloud platforms like Azure, and relay services known as transfer stations.
Is it illegal for Chinese firms to use Claude?
No. The workarounds break Anthropic's terms of service, not US or Chinese law, which is why detection and enforcement fall to the company rather than regulators.
By monitoring account indicators such as computer time zones and usage patterns, and by targeting transfer station services that relay traffic through overseas accounts.
Why does Anthropic block Chinese companies from Claude?
Its terms bar firms majority-owned from restricted regions, citing legal, regulatory, and security risks, including distillation attacks that extract model capabilities into competitor training data.