Anthropic Closes Loopholes Chinese Firms Used to Bypass Claude Ban

Anthropic is tightening controls over who actually uses Claude, after the Financial Times detailed how Chinese tech giants circumvented the ban on China-linked firms that has been in effect since 2025. According to the newspaper's findings, Ant Group and ByteDance used Anthropic's models for months through overseas subsidiaries and employees' private accounts.

This update is designed to prevent such workarounds - Anthropic, statement on tightened access rules

Anthropic's rules have, since 2025, barred Claude access for companies more than 50 percent owned by entities from China, Russia, Iran or North Korea. The ban does not stem from any US or Chinese law but from the company's own terms of service, meaning a violation is not a crime, only a breach of the service's usage terms.

How the workarounds worked

The methods described by the Financial Times were fairly simple. Ant Group registered a company in Singapore and used it to buy corporate Claude accounts for its employees in China, formally satisfying the requirement of being based outside the restricted region. ByteDance took a different route: instead of corporate accounts, engineers bought private Claude subscriptions and connected through a VPN, with the company simply reimbursing the cost through standard expense claims. Other entities relied on overseas companies running on cloud infrastructure from providers such as Microsoft Azure, further obscuring the geographic origin of the queries.

Relay stations under scrutiny

Anthropic said it would go after so-called relay station services, intermediaries that route network traffic through overseas Claude accounts to make it look as though it originates from outside China. The company also plans to monitor accounts for signals such as a user's computer time zone, which can expose a mismatch between a declared location and the actual place of login.

The cost of a hard line

Anthropic's hard-line policy comes at a price. Dario Amodei acknowledged back in February 2026 that cutting off access for firms linked to the Chinese government had cost the company several hundred million dollars in lost revenue. Even so, Anthropic has consistently held its course, citing the need to limit legal, regulatory and security risk, and a desire to keep regimes it describes as authoritarian and hostile from accessing its most advanced models.

The wider US-China rivalry

The case fits into the broader dispute over access to the most powerful AI models between the United States and China, one in which Anthropic has already found itself at the center of attention before, including during the fight over export of its models, which ended when the Trump administration lifted the restrictions. The revelations about Ant Group and ByteDance show, however, that even with official bans in place, Chinese tech giants' appetite for American models remains strong enough that companies are willing to skirt the rules in a variety of creative ways.

For market observers, the episode signals that declarations of full technological isolation between the two powers are hard to enforce in practice, so long as demand for the best language models stays this high and the technical routes around restrictions, such as VPNs or subsidiaries, remain so easy to access.

Sources: Anthropic targets loopholes used by Chinese firms to access Claude, FT reports (investing.com), Anthropic Moves to Block Chinese Firms Using Claude via Offshore Workarounds (banklesstimes.com), Anthropic closes loopholes allowing Chinese access to Claude AI (thenews.com.pk)