Two out of five services that resell access to restricted American AI models accept cryptocurrency, mostly stablecoins, according to research from blockchain analytics firm Chainalysis — a payments layer sitting underneath a market built to evade export controls.
The firm surveyed 100 so-called AI proxy services, which resell access to models including Claude, ChatGPT and Gemini by routing customer requests through intermediary accounts. Users buy credits or top up balances rather than signing up with the developer directly, and the services often offer discounts.
Chainalysis also tracked wallets it identified as potentially connected to China around U.S. frontier model launches. Payments ran 55% above baseline in the first three days after a release, then 7% and 8% above during the following six days. The pattern faded to 21% below baseline on days 12 to 14 before recovering to 10% above baseline on days 18 to 20. The early increase exceeded the pattern recorded across all wallets paying these services.
The firm was careful about what the data shows. Aggregate transaction records cannot establish whether the payments were intended to support model distillation — querying a frontier model to train a competing one — only that spending rose in step with launches.
The report cited Anthropic’s accusation that three China-based laboratories, DeepSeek, Moonshot and MiniMax, used 24,000 fraudulent accounts operating through proxy services to extract capabilities from Claude. That allegation was presented as an illustration of the risk, not as a finding of the on-chain work.
The wallets clustered in China, Japan, Hong Kong, Singapore and South Korea. Chainalysis also found that many proxy providers collect crypto through custodial payment processors rather than accepting funds directly, which separates the user from the operator and complicates the job of any exchange trying to assess its exposure.
The practical upshot is that stablecoins are doing what they were designed to do — moving value across borders quickly, cheaply and without a bank in the middle — in a context where the underlying activity is exactly what U.S. policy is trying to stop. Mapping flows between wallets, payment processors and trading platforms, the firm argued, is the route to identifying the infrastructure behind it.