Anthropic, the San Francisco‑based AI developer, has publicly accused Alibaba of using a large‑scale “distillation attack” to illicitly extract advanced features from its Claude model. The U.S. company says that operators connected to Alibaba carried out almost 29 million exchanges with Claude through thousands of fabricated accounts, claiming it was the largest campaign of its kind.
A formal letter addressed to Senators Tim Scott and Elizabeth Warren notes that the attack targeted Claude’s capacity to handle longer, more complex inquiries and its decision‑making framework. Anthropic argues that such industrial‑scale theft de‑values significant U.S. investment and supplies competitors with a cheap, ready‑made AI stem‑model.
The AI group also cites broader risks, pointing out that similar attacks threaten U.S. military interests. It quotes the Department of Defense’s claims that Alibaba, as well as other Chinese firms like BYD and Baidu, have ties to the Chinese military.
Alibaba has denied all allegations and filed a lawsuit in Washington last week to have its name removed from the Pentagon’s blacklist. The dispute adds to a series of accusations that OpenAI and other U.S. firms have levelled against Chinese companies for employing distillation tactics to replicate American models at lower cost.
Anthropic’s controversies come as the company is primed for a blockbuster IPO, and its own models, such as Mythos, are drawing cybersecurity debates over their ability to identify and exploit system vulnerabilities. The clash emphasizes the urgency of protecting AI research from export‑controlled theft and raises questions about the future of collaborative AI innovation amid a growing digital arms race.
