Anthropic Closes Loopholes Chinese Firms Used to Access Claude

Anthropic is rolling out a new wave of access blocks for companies linked to China. The company revealed that despite earlier bans on sales in China, Chinese tech giants spent months finding ways around the restrictions, using foreign subsidiaries, SIM cards and VPN accounts. Anthropic now says it will tighten controls and actively hunt down these workarounds.

How companies dodged the ban

According to the Financial Times, Ant Group issued employees corporate Claude accounts tied to a Singapore-based subsidiary, a setup that technically sidestepped the ban on direct sales to China. ByteDance engineers, meanwhile, used personal Claude subscriptions paid for by the company, connecting through VPNs to mask their actual location.

The most sophisticated method turned out to be so-called relay stations: servers registered in countries where Anthropic legally sells its services. Such a station would take requests from users in China through a locally accessible website, forward them to Claude via a foreign account or API key, and send the response back, hiding the entire intermediary chain behind cloud infrastructure such as Microsoft Azure.

New detection methods

Anthropic says it will now analyze behavioral signals that give away these workarounds, including the time zone of the connecting user's computer, usage patterns, and the country of issue of the payment card used for the subscription. The company also intends to actively track down and shut down networks of subsidiaries set up solely to gain access to its models.

This extends a policy Anthropic introduced in September 2025, blocking access for companies more than half owned, directly or indirectly, by entities from China, Russia, Iran or North Korea. At the time, Anthropic argued that IP-based blocking alone wasn't enough, since companies could easily get around it through foreign branches.

The cost of the decision

Anthropic CEO Dario Amodei acknowledged earlier, in February 2026, that cutting off access for China had cost the company hundreds of millions of dollars in lost revenue. Anthropic remains the only leading AI lab to restrict sales this way to entities controlled by China, including those registered outside the country.

Amodei frames the policy as an effort to preserve what he calls democratic nations' edge in the race for the most advanced AI models. He argues that companies subject to Chinese law could be compelled to hand over data or cooperate with state security services, and that access to Claude through intermediaries could indirectly support military or intelligence applications.

Market implications

The story gains extra context from Alibaba's recent decision to ban its employees from using Claude Code after discovering a tracking mechanism in the tool. Now Anthropic is actively cutting off access from the other direction, showing just how deeply the US-China tech rivalry has seeped into the daily work of developers and corporations.

For Polish companies and institutions using Claude models, the situation mainly confirms that AI providers are increasingly treating the origin of a client's capital, not just server location, as a security criterion. It's a trend that could spread to other frontier model providers as geopolitical rivalry over artificial intelligence intensifies.

Sources: South China Morning Post (scmp.com), Bankless Times (banklesstimes.com), Anthropic Newsroom (anthropic.com)