Anthropic Urges Congress To Ban AI Distillation By Chinese Critics

Anthropic is asking Congress to strengthen protections against the distillation of the AI ​​model after it said users owned by Alibaba went to great lengths to extract its capabilities from the Claude chatbot.

In a June 10 letter to Senate Banking, Housing, and Urban Affairs Committee Chairman Tim Scott and ranking member Elizabeth Warren, Anthropic reported that employees with Alibaba and the Qwen AI lab made more than 28.8 million transactions with Claude between April 22 and June 5 using approximately 25,000 “fraudulent users,” or not representing fraudulent accounts.

Known as a distillation attackAnthropic said that the work focuses on the mind of Claude agent, software engineering, and long-term planning skills, allowing competitors to produce high-quality models without the cost of training an AI system of limits.

“Beyond its scope, this campaign was remarkable for its vulgarity,” wrote Anthropic. “Alibaba is listed on the New York Stock Exchange, maintains businesses in the United States, and is accountable to US investors and regulators.”

Anthropic said the campaign went beyond the technical implications, casting large-scale distillation as a national security issue that could advance China’s military and cyber AI capabilities and limit the United States’ technological superiority.

The letter comes as Washington steps up efforts to protect US AI leadership. Earlier this month, President Donald Trump he signed A major law that is expanding AI-driven cybersecurity efforts later delay a measure related to anxiety could weaken America’s competitiveness against China.

“When PRC laboratories remove this capability from US models, they capture the returns on the American economy without the costs or risks associated with traditional AI studies,” Anthropic wrote. “This is changing the economic logic that underwrites America’s AI leadership, turning billions of dollars of research and development, computing, and other US businesses into a boon for competitors.”

Anthropic urged lawmakers to increase intelligence sharing between AI developers and the US government, clarify antitrust laws to allow AI companies to share information about distillation attacks, strengthen foreign control of advanced AI chips and calculations, close the doors that allow Chinese companies to enter foreign data centers, and impose sanctions on companies responsible for removing large samples.

An Anthropic spokeswoman declined to comment on the letter, but said Decrypt“We believe that addressing the threat of illegal distillation requires cooperation between government and industry, and we will continue to work with Congress and the administration to maintain American AI leadership.”

The letter also builds on Anthropic’s what to say in February that Chinese AI developers DeepSeek, Moonshot AI, and MiniMax had created more than 16 million copies of Claude using nearly 24,000 fake accounts.

The claims were criticized by observers who said AI companies rely on the same techniques to train their systems. Anthropic has argued that simple distillation is a legitimate method of producing small, cheap samples, while the illegal extraction of the sample’s borderline art through fraud violates its rules.

The debate over distillation has been heated in recent months. In April, Elon Musk he testified in court that xAI “used” OpenAI samples to train Grok, proving that distillation is a standard practice in the industry – although the industry disputes where legal training ends and removing illegal samples begins.