The Pulse
Chinese Resellers Route Claude Into a Gray Market
A gray market of Chinese resellers is selling access to Anthropic’s Claude through pooled accounts and overseas servers. The Information’s October 2 report describes how sellers acquire accounts, evade bans and expose some customers’ data t
AI.info Team · Written with AI ·
“That's only part of the entire picture.”
Jing Yang, Asia Bureau Chief, The Information
Yang was describing the scale of China’s market for unauthorized access to U.S. AI models. The Information reported on October 2 that resellers are supplying Chinese customers with Anthropic’s Claude through networks of accounts and overseas servers, despite Anthropic not offering its services in China.
In Beijing’s Haidian District, six of roughly 30 tenants in one office building sold access to Claude and other U.S. models, according to an occupant interviewed by the newspaper. The reseller estimated that thousands, possibly tens of thousands, of similar firms operate across China. The estimate comes from a market participant, not a verified count.
Overseas servers connect pools of accounts
The resellers’ setup resembles a model-aggregation service, but the accounts behind it are not authorized distribution channels. Operators collect individual Claude accounts and route customer requests through a server, often outside China, that acts as a reverse proxy. Customers use the intermediary rather than connecting directly to Anthropic.
A Shanghai reseller told The Information that he and associates register accounts with email addresses accumulated during earlier work in crypto. He also applies for accounts directly with Anthropic and through coding platforms that offer Claude. The seller said he typically buys Anthropic’s Claude Code Max 20x subscription, priced at $200 a month, and uses payment cards funded with USDT, a stablecoin.
Resellers also advertise services on Chinese social media and GitHub, while negotiating transactions through messaging apps including Telegram. Demand rose early in 2026 amid a surge of interest in OpenClaw, an open-source AI agent platform, the newspaper reported. A growing number of sellers has squeezed some operators’ margins.
Account bans push sellers to find new routes
Anthropic sometimes bans accounts after detecting unusual activity, such as unusually heavy usage, the Shanghai reseller said. He told The Information that he has sought refunds when accounts are suspended and that Anthropic has at times blocked payment cards in batches. The company has also begun asking some users to verify their identities when they log in, according to the report.
As enforcement tightens, some Chinese AI labs are looking beyond pools of individual accounts. Resellers and employees at Chinese model developers told the newspaper that companies have used overseas shell firms to seek enterprise access in regions where Anthropic operates. Some startups outside China that already hold enterprise plans also resell unused capacity to Chinese labs, those sources said.
The report also describes a risk for smaller customers: a reseller’s proxy server can see the requests users send to a model. Some sellers try to monetize that data by selling it to model developers, according to resellers interviewed by The Information. The report says larger AI labs may have safeguards against exposing data to an intermediary, but individual developers and smaller firms may not.
Distillation claims remain attributed to Anthropic
Anthropic has said that proxy services help Chinese AI labs get access to its models and collect exchanges that can be used to train other systems. In its September threat-intelligence report, the company said some resellers save conversations without users’ knowledge or consent, and that labs can buy those transcripts. Those are Anthropic’s findings and accusations, not independent determinations about every reseller or Chinese model developer.
Model distillation—training one model using outputs from another—is used across the industry, including by U.S. AI companies developing smaller versions of their own models. The dispute centers on whether companies obtain proprietary model outputs through unauthorized access and whether they do so without users’ knowledge. Chinese AI companies named in Anthropic’s earlier allegations did not respond to requests for comment or declined to comment for The Information’s report.
For Anthropic, the gray market complicates a restriction that appears simple on paper: Claude is not sold commercially in China, yet account pools, payment intermediaries and overseas proxies can move requests across that boundary. For users buying access through those networks, the immediate tradeoff is just as concrete: a reseller may keep Claude available, but it can also sit between the customer and every prompt they send.