Anthropic has moved to close API access pathways that Chinese companies have exploited to circumvent export controls on its Claude artificial intelligence model, the Financial Times reported.
The AI safety startup implemented new restrictions targeting loopholes in its terms of service that allowed firms based in restricted jurisdictions to access Claude through intermediary accounts and reseller networks. These workarounds enabled Chinese organizations to bypass U.S. export regulations designed to prevent advanced AI technology transfer to geopolitical rivals.
The action reflects growing tension between AI developers and enforcement of Biden administration export controls on frontier AI models. The Commerce Department imposed restrictions in October 2023 specifically targeting countries deemed national security concerns, including China. Anthropic's restrictions aim to prevent sophisticated routing of API calls through third-party vendors and shell corporate structures.
"We've identified that some customers may be using Claude through third parties to circumvent our export controls," Anthropic stated, according to the report. The company notified affected users and implemented verification protocols to confirm the actual location and ownership structure of API account holders.
This move aligns Anthropic with similar enforcement actions by competitors. OpenAI has faced persistent challenges detecting and stopping Chinese organizations from accessing ChatGPT through proxy services and business partnerships. Both companies face a technical and compliance dilemma: AI APIs require international connectivity to function, yet they must prevent circumvention schemes.
Anthropic's enforcement targets several common loopholes. Resellers in permissible jurisdictions often route API access to end customers in restricted regions. Some Chinese firms created shell companies in Singapore, Hong Kong, and other intermediary jurisdictions to establish seemingly legitimate accounts. Payment methods and billing addresses provided additional cover for these arrangements.
The Financial Times investigation identified multiple Chinese AI research institutes and tech companies benefiting from Claude access despite export restrictions. These organizations used the model for research, competitive analysis, and internal product development.
Anthropic joins a broader push by U.S. AI firms to tighten compliance enforcement. The company operates under Commerce Department regulations that allow limited exceptions for research and humanitarian purposes, but these carve-outs have themselves become vectors for circumvention schemes.
Investors monitoring AI software companies should track enforcement capacity and compliance costs as ongoing regulatory pressures on frontier AI exporters intensify. Export control enforcement affects competitive positioning and liability exposure for publicly traded AI platforms.