Anthropic is moving to shut loopholes that have enabled Chinese companies to bypass its strict restrictions on unauthorized use of Claude AI tools in China. The AI company has implemented some of the toughest bans among US tech firms, requiring user verification and blocking payments from Chinese banks.

Chinese firms including Ant Financial have accessed Claude through various workarounds, including cloud providers and overseas subsidiaries. Ant reportedly provided employees with corporate Claude accounts accessed via its Singapore-based entity's intranet. Meanwhile, ByteDance allows engineers to expense personal Claude subscriptions accessed through VPNs, according to five employees.

These access methods don't violate US or Chinese law but breach Anthropic's terms of service, which prohibit Chinese companies and their foreign subsidiaries from using Claude models. The workarounds highlight continued demand for leading US AI products among Chinese engineers despite growing restrictions and competitive domestic alternatives.

Claude's coding tools are particularly sought after because outputs can be used for "distillation" - training smaller models to mimic more capable ones. Microsoft admitted selling API access to Chinese companies with Singapore entities, allowing mainland engineers to use Claude through internal networks. Anthropic said it continuously updates detection systems to identify and ban violating accounts.