Alibaba Named in Anthropic's Letter on AI Model Theft

Anthropic is asking the U.S. government to step in against Chinese companies it says are copying its AI models, and it named Alibaba Group Holding as one of the attackers.

The American AI company sent a letter to several U.S. senators and White House officials, Bloomberg reported Wednesday, citing a copy of the document. An Anthropic spokesperson declined to discuss the letter with Bloomberg but said there is a need for "coordinated action between government and industry" to combat distillation, a process where developers use results from another AI model to train their own at a much lower cost.

Alibaba declined Bloomberg's request for comment. The company did not immediately reply to PYMNTS' request for comment.

According to the Bloomberg report, Anthropic said in its letter that Alibaba's Qwen AI lab used 25,000 fraudulent accounts to illicitly access its Claude AI model, engage in 28.8 million exchanges with the model, and use the information to develop rival chatbots through adversarial distillation. The company said AI models developed this way lack the guardrails Anthropic puts on its own models, and that distillation poses a national security threat because it could help China reduce America's lead in AI.

Anthropic also asked the senators and officials to clarify antitrust guidelines so U.S. companies can share information about distillation, support export controls on advanced AI chips, and penalize firms that use the technique.

Anthropic said in a February blog post that Chinese AI labs DeepSeek, MiniMax and Moonshot AI had used distillation and Claude's outputs to train their own models. "These campaigns are growing in intensity and sophistication," Anthropic said in the post. "The window to act is narrow, and the threat extends beyond any single company or region."

Google Threat Intelligence Group said in a Feb. 12 blog post that it had seen a growing incidence of distillation attacks or "model extraction attacks" and that a coding model "could be targeted by an adversary wishing to replicate capabilities in an environment without guardrails."

Alibaba's BABA stock page carries an Alpha Score of 45 out of 100, a Mixed label in the Consumer Discretionary sector. The score reflects the competing pressures of its core e-commerce business and the regulatory and competitive risks around its AI operations.