ME AI message, June 25 (UTC+8): According to Bloomberg, Anthropic, in a letter to U.S. Senators and White House officials, accused Alibaba’s Qwen Lab of using nearly 25,000 fraudulent accounts to conduct large-scale unauthorized access to Claude. The letter revealed that between April and June 2026, the associated accounts interacted with Claude up to 28.8 million times, focusing primarily on core capabilities such as software engineering and agent reasoning.
Anthropic warns that Alibaba and Chinese laboratories are systematically and illegally using outputs from U.S. frontier models to train competing models at extremely low R&D costs. Adversarial distillation is harvesting U.S. AI capabilities at an industrial scale, and the resulting models often lack safety safeguards. To counter this attack, Anthropic urges the U.S. government to clarify antitrust guidelines to eliminate legal concerns preventing U.S. companies from sharing intelligence on distillation attacks, and proposes imposing sanctions on entities that exploit distillation techniques to steal model capabilities.
In response to the joint boycott by major U.S. companies, bipartisan U.S. senators are planning to introduce an amendment to the National Defense Authorization Act to blacklist Chinese companies that use illegally accessed outputs of U.S. AI models to train competing models.
Currently, Anthropic, OpenAI, and Google have established a collaboration mechanism to share distilled intelligence on violations of terms of use.(Source: Bloomberg)