US AI firm Anthropic accuses Alibaba of using fake accounts to steal AI model data

US artificial intelligence startup Anthropic has accused Chinese tech giant Alibaba of illicitly leveraging its AI model, Claude, to develop competing technologies. The accusation centers on the alleged use of tens of thousands of fake accounts to access Claude's core functionalities.

Anthropic detailed its claims in a letter to US senators and White House officials, stating that operators linked to Alibaba's Qwen Lab created approximately 25,000 fake accounts between April and June. These accounts reportedly engaged with Claude over 28.8 million times, with a particular focus on sensitive areas like software engineering and AI agent reasoning.

The core of the dispute lies in a technique known as "distillation." This method allows developers to train new AI models more cheaply by using the outputs of larger, existing models as training data. While common for developing smaller models, leading AI firms argue that using their proprietary models for distillation without permission violates their terms of service.

Anthropic described the alleged actions as the largest-scale attempt by a Chinese company to harvest US AI technology. The company urged the US government to strengthen regulations against such practices, which it characterized as "industrial-scale collection" of cutting-edge AI research. Alibaba has declined to comment on the allegations, and the report caused a dip in Alibaba's stock price.

“

This distillation attack is an industrial-scale collection of the capabilities of US-based leading AI labs, repackaged as their own.