How China's Underground Economy Keeps Accessing Claude Despite Bans
Anthropic built a wall. China found a ladder.
That's the gist of a deep dive from Wired on the thriving underground economy for Claude access in China. Despite escalating blocks from Anthropic — VPN-detecting IP checks, API key restrictions, the works — users in China keep finding ways in. And they're paying for the privilege.
The Transfer Station Model
The most common workaround: "transfer station" sites that buy Claude API tokens abroad (typically through intermediary markets), then resell them to Chinese users at a markup. These services handle the geolocation spoofing, managing the technical cat-and-mouse with Anthropic's trust & safety team.
Telegram plays a big role here. Vendors sell fake identities, pre-configured API keys, and even "guaranteed" access passes — all denominated in USDT (Tether), making the transactions hard to trace.
Why Claude?
The obvious question: why go through all this trouble when Chinese AI companies exist? The answer is quality. Users consistently rate Claude's reasoning, coding output, and English-language capabilities above domestic alternatives like Zhipu (GLM-5.2) and Baidu's Ernie.
One source quoted in the piece put it bluntly: the best models are still American, and that hasn't changed even as Chinese labs close the gap.
The Irony
Anthropic's export restrictions were designed to prevent misuse of its most capable models (Mythos, Fable 5). Instead, they've created a gray market that benefits no one except proxy operators. Meanwhile, Anthropic is reportedly close to restoring Fable 5 access for non-Chinese users — but the damage to its brand in China may already be done.
The broader lesson: building walls around intelligence is harder than building the intelligence itself. When people want something badly enough, they'll find a way.