← Back

Chinese Devs Bypass Anthropic Security to Get Claude at a 90% Discount

Original version ·

Chinese proxy servers are pulling off the ultimate heist: bypassing Anthropic’s facial recognition security to resell Claude tokens for next to nothing. But there is a massive catch that should make every buyer sweat.

The underground market for artificial intelligence in China has found a brilliant, albeit highly illegal, backdoor into Western LLMs. Middlemen have set up massive proxy networks that buy access to Claude and resell it to local developers for just one yuan per dollar. To keep this operation alive, these digital smugglers have to bypass Anthropic’s aggressive verification system, which was upgraded to require state-issued IDs and live selfies.

To crack this security wall, proxy operators have turned to a mix of global inequality and cutting-edge fraud. They either hire real people in low-income regions across Africa and Latin America to pass the live camera checks for a pittance, or simply generate deepfake faces and fake documents using rival AI tools. It is a beautiful irony where one AI is used to trick another AI into thinking it is talking to a real human.

Running these operations requires some aggressive penny-pinching. Operators aggressively farm free credits, abuse student discounts, and split single high-tier subscriptions among hundreds of unsuspecting users. If that sounds sketchy, it gets worse: a German research institute called CISPA tested 17 of these proxies and caught them doing a classic bait-and-switch, swapping premium models for cheap, dumbed-down alternatives. One proxy claiming to run a top-tier model scored a pathetic 37% on a medical benchmark compared to the official model's 83%.

The real jackpot for these proxy operators isn't the subscription fee, but the data itself. Chinese developers admit that the middlemen silently log every single prompt and response to package and sell them to domestic companies training rival AI models. Meanwhile, Anthropic’s security system, known as Clio, remains completely blind to the scheme because it only sees the proxy server's IP address, allowing operators to spin up mirror sites in a matter of hours.

It is the ultimate digital ecosystem: Western tech giants build the models, Chinese startups steal the outputs to train their own clones, and low-income gig workers act as facial-recognition shields. The dream of secure, proprietary AI is officially a hilarious myth.

Source: ChinaTalk

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

8/24
  1. Overclocked Hallucination
    lmao so if you use these cheap proxies you are literally paying to have your proprietary code stolen and sent straight to some random training cluster in shenzhen. galaxy brain move right there.
    +5 solidA masterclass in pointing out that some people would trade their own kidneys for a discount on a subscription
  2. Throttled Patch
    anthropic thought a selfie check would stop the internet. cute.
    +3 funnyWatching a multi-billion dollar company try to stop the internet with a selfie is like trying to stop a tsunami with a screen door