Chinese engineers reportedly found secret routes to Claude, Anthropic is shutting them down

Anthropic is stepping up efforts to prevent companies in China from accessing its AI models through indirect channels, following reports that several major Chinese firms have been using overseas entities and cloud infrastructure to bypass the company's restrictions.

According to a report by the Financial Times, Chinese technology groups, including Ant Financial, have continued using Anthropic's AI services despite the company's policy prohibiting access by Chinese organisations and entities under their control. The reported workarounds rely on overseas subsidiaries, corporate networks and cloud platforms rather than direct access from mainland China.

The development underscores the increasing challenge facing US AI companies as they attempt to enforce regional access restrictions while demand for advanced AI models continues to grow globally.

Anthropic has adopted one of the toughest China policies among leading AI developers. Unlike some competitors, it requires user verification, blocks payments from Chinese banks and explicitly prohibits access by companies based in China or foreign subsidiaries controlled by them.

More from Tech

Overseas entities and cloud infrastructure emerge as access routes

The Financial Times reported that Ant Financial enabled employees to use Claude through corporate accounts connected to its Singapore-based entity, allowing staff to access the service through the company's internal network.

The report also said ByteDance, the parent company of TikTok, did not directly provide Claude access to employees. Instead, it reportedly introduced a reimbursement programme allowing engineers to expense personal Claude subscriptions, with users relying on virtual private networks (VPNs) to connect to the service.

Although these methods do not violate either US or Chinese law, they are said to breach Anthropic's terms of service.

Other companies have also used overseas subsidiaries to access Claude through Microsoft's Azure cloud platform, reports FT. Under this arrangement, engineers working in mainland China could reportedly use Claude via their employers' overseas corporate infrastructure.

Anthropic strengthens enforcement as AI rivalry intensifies

Anthropic confirmed it is continuously updating its detection systems to identify attempts to evade its geographical restrictions.

"We explicitly prohibit accessing or facilitating access to Claude in unsupported regions, including China," the company said.

"Anthropic is the only frontier AI company that restricts sales to PRC-controlled companies, including subsidiaries incorporated outside China."

The company added that it "enforces this policy through continuous, evolving detection systems, working alongside our partners to identify and ban accounts that violate our policies".

People familiar with the matter told the Financial Times that Anthropic has also targeted so-called "transfer station" services. These intermediaries relay prompts from users in mainland China through Claude accounts registered overseas before sending responses back to the original user.

However, the report noted that many larger Chinese AI companies avoid relying on such services because of concerns that operators may retain or resell prompts. Executives reportedly fear sensitive requests could reveal how they are developing their own AI models or training techniques.

Anthropic has also deployed technical methods to detect unauthorised use. In previous enforcement efforts, the company reportedly used its own coding assistant, Claude Code, to identify indicators such as a computer's time zone that suggested a user was operating from mainland China.

Despite the restrictions, Claude remains highly sought after by Chinese software engineers and AI start-ups. According to the Financial Times, the company's coding tools are particularly valued because their outputs can support model "distillation", a process in which smaller AI systems are trained to replicate the behaviour of more capable models.

China's own regulations prohibit companies from using foreign AI models hosted overseas to build consumer-facing applications due to cross-border data transfer rules. However, those rules do not prevent Chinese AI laboratories from using overseas models for internal research and development, leaving continued demand for leading US AI systems even as access becomes increasingly difficult.