News / US / cnbc
Anthropic Alleges Chinese AI Labs Secretly Harvested Claude Data for Model Training
Technology Services · Internet Software/Services · cnbc · 2026-09-11
BABA
Anthropic reports that major Chinese AI firms, including Alibaba and Moonshot, illicitly used millions of Claude interactions to train their own proprietary models.
What Happened
Unauthorized Data Harvesting: Anthropic has identified large-scale efforts by Chinese AI labs to scrape data from its Claude model to improve their own systems. This practice, termed illicit distillation, involves using outputs from a more advanced model to replicate its capabilities without proper authorization or licensing.
Scope of the Operation: The report highlights significant activity from Alibaba, Moonshot, and DeepSeek, with Alibaba alone accounting for over 151 million exchanges between May and July. These firms allegedly utilized thousands of fraudulent accounts to bypass security measures and funnel user requests through Claude's interface.
Privacy and Security Risks: Anthropic expressed concern that these unauthorized exchanges included sensitive information from individual users and multinational corporations. The company noted that these actions likely violate privacy regulations and the specific terms of service governing the use of its AI platform.
Operational Tactics: Moonshot and DeepSeek reportedly routed customer queries to Claude without informing their users, effectively masquerading Claude's responses as their own. Anthropic has since taken steps to disrupt these operations, which spanned various activities including cyber surveillance and model distillation.