Anthropic has accused Chinese e-commerce and tech giant Alibaba of running a large-scale effort to extract capabilities from its Claude AI model. According to the company, the operation relied on tens of thousands of fake accounts that sent millions of queries to the system in less than two months.
In a June 10 letter to Senators Tim Scott and Elizabeth Warren, the San Francisco-based AI startup alleged that accounts linked to Alibaba’s Qwen AI team initiated more than 28.8 million conversations with Claude between April 22 and June 5. Anthropic described the activity as the largest corporate espionage campaign it has recorded.
Anthropic says the campaign was a coordinated “distillation attack,” or model extraction attack. The idea is to repeatedly query a competitor’s AI through its public API, then use the responses to improve another model. While each request may look like a regular user interaction, the process happens on a huge scale with millions of carefully planned prompts.
Anthropic says the goal was to help Alibaba’s Qwen model learn the capabilities of its newest AI model, Mythos Preview. According to the company, Mythos Preview is a big step up from older models like Claude Opus, particularly when it comes to coding and digital security tasks.
Anthropic went a step further by claiming the Chinese government was involved in the attacks, saying they were part of Beijing’s larger ambition to become the global leader in AI and machine learning. The company warned that if these efforts succeed, they could become a major risk for the United States and its allies.
The letter was addressed to Senators Tim Scott and Elizabeth Warren, the chair and ranking member of the U.S. Senate Committee on Banking, Housing, and Urban Affairs. The committee held a two-day hearing on artificial intelligence on June 11 and 12 under the theme, “AI and the American Dream: Promoting Innovation, Affordability, and American Dominance.”
Two days after Anthropic sent the letter, the U.S. Commerce Department imposed strict controls on the company’s Mythos and Fable AI models, citing concerns that China and other blacklisted countries could use their advanced capabilities against U.S. interests. Anthropic has since suspended public access to both models worldwide while it awaits further guidance from federal authorities.
Anthropic has made similar allegations before. In February, the company said it identified a separate campaign involving operators connected to three Chinese AI startups, including DeepSeek, that allegedly coordinated distillation attacks to enhance their own models.
Maybe you’d like some other interesting articles?