News
Chinese Proxy Services Bypass U.S. AI Export Controls Using Crypto
Chinese proxy services exploit crypto payments to access U.S. AI models, undermining export controls and fueling a shadow market for advanced AI capabilities.
A shadow market of proxy services is enabling Chinese users to bypass U.S. export controls on advanced AI models, according to research by Chainalysis. These services provide access to restricted frontier models, such as ChatGPT and Claude, allowing Chinese companies to distill advanced capabilities into their own systems at a fraction of the training cost. The report highlights the role of cryptocurrency in facilitating this market, with 40% of surveyed proxy services accepting crypto payments, predominantly stablecoins.
How Proxy Services Operate
Proxy services act as intermediaries, routing user queries into restricted AI models while concealing the origin of the requests. Payments for these services are made through fiat currency, including platforms like WeChat and AliPay, as well as cryptocurrencies. By using crypto, these services can bypass state capital controls and maintain off-grid access for sensitive users. Chainalysis found that many proxy services rely on custodial cryptocurrency payment processors, adding a layer of separation between users and providers.
The ecosystem has evolved into a structured economy with a segmented supply chain. Vendors acquire access to AI models through methods such as account farming and model swapping, while proxy services handle downstream operations like pooling access and selling subscriptions. This layered structure makes enforcement and disruption of the market particularly challenging.
Impact on U.S. National Security
The unauthorized access facilitated by proxy services undermines U.S. national security-focused restrictions. Chinese AI companies use these services to distill outputs from advanced models, improving their own systems without incurring the high costs of training from scratch. For example, Anthropic has accused Chinese labs such as DeepSeek, Moonshot, and MiniMax of using 24,000 fraudulent accounts to extract capabilities from its Claude model. These illicitly distilled models often lack safeguards, posing risks to critical infrastructure and other domains.
The National Security Agency recently warned that such unauthorized models could spread dangerous capabilities. The shadow market’s reliance on crypto payments further complicates enforcement, as funds linked to proxy services can reach regulated platforms without clear ties to the original transactions.
Crypto Activity Spikes After AI Model Releases
Chainalysis observed increased activity from suspected Chinese-linked cryptocurrency wallets following the release of new U.S. frontier AI models. This pattern suggests efforts to quickly access and distill new capabilities. While the data does not confirm causation, it aligns with the use of crypto as a payment rail in the proxy economy. Wallet activity was concentrated in China and nearby regions, including Japan, Hong Kong, Singapore, and South Korea, indicating the use of regional proxy corridors.
Challenges in Enforcement
The mature shadow economy around AI proxy services presents significant enforcement challenges. Blocking direct access to frontier models is insufficient when a multi-tiered supply chain exists to resell access. Blockchain intelligence tools could play a critical role in tracing financial flows within this ecosystem, potentially exposing links between proxy services and their users. However, the adaptability of the market, with new vendors stepping in when others are disrupted, complicates efforts to dismantle it entirely.