Anthropic and China: The Escalating Battle Over Claude Code

The release of Claude Code has ignited a complex geopolitical and corporate standoff, revealing deep-seated tensions between Western AI developers and Chinese tech giants. As Anthropic attempts to enforce strict usage restrictions, a cycle of bans, workarounds, and surveillance-like safeguards is reshaping the landscape of global AI development.

Anthropic’s Enforcement vs. Corporate Workarounds

Anthropic has taken a firm stance to prevent Chinese entities from accessing its advanced coding capabilities. According to reports from the Financial Times, the company’s terms of service explicitly prohibit sales to any organizations controlled by the Chinese government. This move is designed to protect intellectual property and comply with shifting regulatory pressures.

However, enforcement remains a massive challenge in a hyper-connected digital economy. Major Chinese players, including Ant Financial and ByteDance, are reportedly bypassing these restrictions. These companies are leveraging sophisticated methods such as utilizing third-party cloud services, routing traffic through overseas subsidiaries in Singapore, or employing VPNs to mask their identity and origin. This cat-and-mouse game highlights the difficulty of enforcing geographic boundaries in a software-defined world.

Alibaba’s Countermove: The Internal Ban

While Anthropic is trying to keep Chinese companies out, Chinese tech giants are increasingly locking their own doors. The Information reports that Alibaba has taken the drastic step of banning its employees from using Claude Code. Furthermore, the company is reportedly requiring staff to delete all existing Claude models from their systems.

This defensive maneuver follows revelations regarding "hidden code" within Claude Code. There were reports that the tool contained mechanisms capable of flagging users who were physically located in China or were associated with Chinese research laboratories. Anthropic’s Thariq Shihipar addressed these concerns, characterizing the code as a temporary experiment conducted in March aimed at preventing account abuse and "distillation"—the process of using a larger model's outputs to train a smaller, cheaper model. Anthropic maintains that these experimental measures have since been replaced by more robust safeguards.

The War Against Model Distillation

At the heart of this friction is the issue of model distillation. Anthropic has been vocal in its accusations against several prominent Chinese AI firms, including Alibaba, DeepSeek, Moonshot AI, and MiniMax. The core allegation is that these companies are using Claude’s high-quality outputs to train their own proprietary models, essentially "stealing" the reasoning capabilities of Anthropic’s frontier models to accelerate their own development.

This conflict represents a broader trend in the AI industry: the struggle to protect the massive capital investments required to build LLMs. As the distinction between "using" an AI and "training on" an AI blurs, the legal and technical battlegrounds between the US and China will likely become the defining feature of the next era of artificial intelligence.

Key Takeaways

• Circumvention Strategies: Despite Anthropic’s explicit bans on China-controlled entities, firms like ByteDance and Ant Financial are bypassing restrictions via Singaporean subsidiaries and cloud workarounds.
• Corporate Defensive Bans: Alibaba has prohibited the use of Claude Code among its workforce following reports of code meant to identify Chinese-linked users.
• The Distillation Conflict: A major driver of this tension is Anthropic's battle against "distillation," where Chinese firms like DeepSeek and Moonshot AI are accused of using Claude to train their own models.