Chinese grey market resells Anthropic Claude API access at 90% discount
Updated
Updated · Tom's Hardware · May 9
Chinese grey market resells Anthropic Claude API access at 90% discount
5 articles · Updated · Tom's Hardware · May 9
Researcher Zilan Qian said proxies on GitHub, Taobao and Telegram sell access for 10% of official prices, using stolen credentials and recruiting people abroad to pass identity checks.
A CISPA audit of 17 services found widespread model substitution, while operators reportedly capture prompts and outputs, exposing source code and generating valuable training data for distillation.
The findings echo White House and Anthropic warnings about Chinese proxy-account abuse; Anthropic previously linked about 24,000 fraudulent accounts to Chinese labs despite tighter verification and access restrictions.
As China's grey market turns stolen data into rival AI, are US firms fueling their own competition?
When AI services secretly steal code and swap models, how can anyone trust the AI they use?
With deepfakes defeating identity checks, are security measures for frontier AI models already obsolete?
24,000 Fraudulent Accounts, $10 Billion at Stake: Anthropic’s Struggle to Secure Claude Amid US-China AI Tensions
Overview
Anthropic has stepped up its fight against the grey market for its advanced AI models by deploying robust technical measures and policy advocacy. The company focuses on preventing distillation attacks, which allow foreign entities to strip away the guardrails of its AI models. These attacks pose serious risks, such as enabling the creation of bioweapons or surveillance tools, and threaten the competitive edge that export controls are meant to protect. By combining identity scoping, sandboxing, and cutting off known illicit access points, Anthropic aims to safeguard its technology and address national security concerns.