Sep 11, Kathmandu - In a startling revelation, AI company Anthropic claims that China's tech giant Alibaba has conducted the largest unauthorized data extraction campaign from its AI model, ‘Claude’. 

According to Anthropic, between May and July 2026, accounts linked to Alibaba interacted with Claude over 15 million times, raising serious concerns about data security and intellectual property theft.

Anthropic alleges that individuals associated with Alibaba secretly stole vast amounts of data from Claude to train their own AI model, ‘Quen’. This incident marks the largest known data breach of its kind to date.

The company revealed that some accounts posed up to 3 million questions in a single day, using over 3,500 fake accounts to carry out the operation. These accounts repeatedly asked Claude the same questions to understand how it processes and derives answers (chain-of-thought reasoning), and then used those responses to train Quen.

After blocking these accounts, Anthropic reports that Alibaba launched a new wave of activities from other accounts. Surprisingly, these accounts were also linked to other Chinese firms like DeepSik and Siomi.

Since February 2026, Anthropic has uncovered similar activities involving seven Chinese companies, including DeepSik, Moonshot AI, Xiaomi, Zhipu AI, Sensetime, and Minimax, engaging in comparable data extraction efforts.

In response to the escalation, Anthropic announced plans to improve its fake account detection system and implement mandatory identity verification for suspicious accounts arriving from China, Russia, and Iran.

Alibaba has yet to comment on the allegations.