Artificial intelligence company Anthropic has officially accused Chinese technology giant Alibaba of improperly using its Claude AI models to train Alibaba’s own artificial intelligence systems. According to Anthropic, the alleged activity constitutes what is known in the AI industry as a “distillation attack,” a practice in which a smaller or less capable AI model learns from the outputs of a more advanced model rather than being trained entirely from scratch.
Model distillation itself is a common machine learning technique when conducted with authorization. However, Anthropic claims that Alibaba obtained access to Claude’s capabilities through deceptive means and used the resulting data to improve its own AI models without permission. The company outlined its concerns in a letter dated June 10, 2026, addressed to U.S. Senate Banking Committee Chairman Tim Scott and Ranking Member Elizabeth Warren.
According to the allegations, Alibaba’s AI division, known for developing the Qwen family of models, created approximately 25,000 fraudulent accounts between April 22 and June 5, 2026. Through these accounts, Anthropic claims that more than 28.8 million interactions were conducted with Claude. The company believes these exchanges were part of a large-scale adversarial distillation effort designed to extract knowledge, behavioral patterns, and response capabilities from Claude and transfer them into Alibaba’s own systems.
If proven true, such actions could have significant implications for the AI industry. Training state-of-the-art generative AI models typically requires enormous investments in research, engineering talent, computing infrastructure, and energy resources. By leveraging the outputs of an existing advanced model, a competing company could potentially reduce development costs dramatically while accelerating the performance of its own products.
Anthropic argues that this practice enables foreign AI laboratories to replicate capabilities developed by U.S. companies with comparatively limited research and development expenditure. The company views this as not only a commercial concern but also a strategic issue with broader implications for technological competitiveness and intellectual property protection.
However, experts note that the U.S. Senate Banking Committee has limited direct authority in such matters. While lawmakers may recommend sanctions or regulatory actions against foreign entities, enforcing such measures against major overseas corporations can be difficult and politically complex. As a result, any immediate consequences for Alibaba remain uncertain.
The controversy first gained public attention through reporting by Bloomberg, which investigated the allegations and highlighted growing concerns surrounding AI model security and unauthorized knowledge extraction. The case has also emerged amid broader efforts by the Trump administration to tighten restrictions on foreign access to sensitive American technologies. In response to perceived AI-related threats, the administration has issued executive orders aimed at preventing foreign organizations from exploiting U.S. technological resources, with similar restrictions reportedly being considered for upcoming AI systems such as Fable 5 and Mythos 5.
As AI competition intensifies globally, disputes over model distillation, intellectual property, and technological security are expected to become increasingly important issues for both governments and technology companies.
Join our LinkedIn group Information Security Community!
