Anthropic is taking steps to close loopholes that have enabled users and organizations in China to gain access to its Claude AI models, according to a report from the Financial Times. The move reflects growing pressure on American AI companies to enforce geographic and geopolitical restrictions on their most capable systems, as concerns mount in Washington about advanced AI technology reaching adversarial states.

A Pattern of Unauthorized Access

The announcement is not an isolated event. Anthropic has previously accused Alibaba of running a coordinated campaign to access Claude in violation of its usage policies. Those allegations detailed a systematic effort to route queries through third-party services and API resellers that were not subject to the same restrictions applied directly to users in China. The latest enforcement push appears designed to address exactly these kinds of indirect access routes, which have proven difficult to police under previous policy frameworks.

Key Facts

  • Anthropic is updating its policies and technical controls to block indirect Chinese access to Claude.
  • Previous restrictions were reportedly circumvented through third-party API resellers and intermediary services.
  • The move follows public accusations against Alibaba and other Chinese entities over unauthorized use.
  • U.S. export control concerns are a significant backdrop to Anthropic's enforcement actions.
  • The changes affect how Anthropic vets and monitors its API partners and downstream customers.

Closing these loopholes requires more than updating terms of service. Anthropic must also scrutinize its network of API partners, since many unauthorized users reached Claude not through direct accounts but through legitimate resellers who were not adequately screening their own customers. That makes enforcement a layered problem, combining legal, contractual, and technical measures. Earlier reporting detailed how Anthropic accused Alibaba specifically of illicit access, suggesting the company has been gathering evidence of these patterns for some time before moving toward broader structural fixes.

Anthropic is committed to ensuring its technology is not accessed in ways that violate its policies or applicable law, and is taking concrete steps to address known gaps in its enforcement mechanisms.Anthropic spokesperson, via Financial Times

Broader Context: AI, Trade, and National Security

The enforcement effort sits against a backdrop of intensifying U.S. government scrutiny of AI exports. Federal regulators and lawmakers have been pushing AI developers to implement more rigorous controls, particularly for foundation models capable of aiding in scientific research, defense applications, or surveillance. Claude models have demonstrated advanced problem-solving capabilities, which makes them a potential target for state-backed actors seeking to leverage cutting-edge AI without access to domestic equivalents.

Anthropic's commercial growth has also added complexity to this picture. The company is closing a funding round that values it near $900 billion, reflecting investor confidence in its technology and business model. That scale brings greater regulatory visibility and responsibility. A company operating at that level cannot afford to be seen as lax on questions of national security compliance, especially as AI policy becomes a more active area of federal attention.

The steps Anthropic is taking now may set a precedent for how other U.S. AI developers handle similar enforcement challenges. API-based access to large language models is inherently difficult to restrict by geography, since requests can be proxied and identities obscured. Building more robust screening into the partner and reseller layer, rather than relying on end-user agreements alone, could become a standard expectation across the industry. For now, Anthropic appears to be ahead of most peers in publicly confronting the problem and articulating a response.