Alibaba Accused of Illegally Mining Data from Anthropic’s Claude to Train Its Own AI
Summary
- Anthropic claims that accounts linked to Alibaba interacted with Claude over 151 million times from May to July 2026, training its AI model Qwen.
- Over 3,500 fake accounts were reportedly created for these interactions; after being blocked, a new set of accounts resumed the activity.
- Anthropic plans to enforce identity verification for suspicious accounts from China, Russia, and Iran, while Alibaba has not responded.
September 12, Kathmandu – Artificial intelligence company Anthropic has accused China’s leading tech giant Alibaba of conducting the largest illicit data mining operation to date, extracting data from Anthropic’s AI model ‘Claude’ to train its own model.
According to Anthropic, Alibaba-associated accounts interacted with Claude more than 151 million times from May through July 2026.
Anthropic stated that individuals connected to the Chinese tech giant Alibaba stole significant amounts of data from Claude, their AI model, using it to train their own AI model named ‘Qwen’.
This incident is described as the largest such case to date.
At times, a single account would submit up to 3 million queries in a day. To facilitate this, more than 3,500 fake accounts were created.
These accounts queried Claude in similar ways, recording the entire reasoning process (chain-of-thought) and the answers provided, which were then used to train the Qwen model.
After Anthropic blocked these accounts, Alibaba resumed its activity using a different group of accounts.
Interestingly, these accounts were not exclusive to Alibaba; other Chinese companies such as DeepSea and Xiaomi also reportedly used them.
In February 2026, following Anthropic’s initial disclosure, similar activities were found involving seven other Chinese companies, including DeepSea, Moonshot AI, Xiaomi, Jipu AI, SenseTime, and Minimax.
With these incidents on the rise, Anthropic has announced plans to improve its fake account detection system and enforce identity verification for suspicious accounts originating from China, Russia, and Iran.
Alibaba has yet to issue any response regarding these allegations.
