Anthropic accuses Alibaba of illicitly training models using Claude data

Anthropic, the creator of the Claude family of large language models, has formally accused Chinese technology giant Alibaba of systematically extracting its model’s capabilities to train competing systems. In a letter sent to U.S. Senators Tim Scott and Elizabeth Warren ahead of an upcoming AI-focused hearing, Anthropic detailed what it describes as a large-scale violation of its terms of service.

According to the company, Alibaba orchestrated an operation involving 25,000 fictitious accounts and nearly 28.8 million data exchanges between April and June 2026. The exchanges, the letter argues, were used to "distill" Claude’s advanced reasoning and language abilities into smaller, more efficient models developed by Alibaba’s Qwen AI lab. This technique, known as model distillation, allows developers to compress the knowledge of a powerful model into a lighter version that requires fewer computational resources to run. While legitimate use cases exist, Anthropic contends that Chinese firms are exploiting the method to replicate its innovations at a fraction of the development time and cost.

The alleged campaign was traced to operators with direct ties to Alibaba Qwen, the AI division behind the company’s open-source language models. Anthropic warns that such practices could enable Chinese developers to rapidly close the technological gap with American counterparts, potentially matching or surpassing the capabilities of leading U.S. models like Mythos Preview. The concern is particularly acute for U.S. policymakers, who have long viewed AI leadership as a critical national security priority.

This is not the first time Anthropic has raised alarms about unauthorized model distillation. Earlier in 2026, the company made similar accusations against three other Chinese AI labs—DeepSeek, Moonshot, and MiniMax—claiming they collectively used 24,000 fraudulent accounts and 16 million data exchanges to train their own models on Claude’s output. The repeated incidents highlight growing tensions in global AI development, where legal compliance and data sovereignty are increasingly becoming flashpoints for geopolitical rivalry.

The escalating dispute reflects broader efforts by both the U.S. and China to gain an edge in artificial intelligence. Washington has intensified export controls to restrict China’s access to advanced semiconductors and chip-manufacturing equipment, while Beijing has retaliated by tightening restrictions on rare earth materials—key components in semiconductor production. These measures underscore the high-stakes nature of AI competition, where data access, computational power, and regulatory environments are as decisive as technical innovation.

As the global AI landscape evolves, the outcome of these allegations could set precedents for how model training data is protected, how international collaborations are conducted, and whether technological progress can be decoupled from geopolitical conflict. With both nations investing heavily in AI infrastructure and talent, the race for supremacy shows no signs of slowing down.