Why Anthropic Accused Alibaba Of Illegally Accessing Claude AI Model

Anthropic I Alibaba: The San-Francisco headquartered Anthropic PBC accused Alibaba Group Holding Ltd of “illicitly” gaining access to Claude's artificial intelligence model via several thousand fake accounts that undermine the US AI developer's decision to keep its products outside China.

The Dario Amodei led company claimed that a campaign linked to the Chinese tech giant's Qwen AI lab targeted Claude’s software engineering and agentic reasoning, according to a letter that the AI startup sent to several US senators and White House officials, citing a Bloomberg report.

The company said it was the largest attempt by Chinese firm to advance on the work of top US labs.

In its letter, Anthropic noted this move involved 28.8 million exchanges with Claude between April and June via nearly 25,000 fraudulent accounts, according to people familiar with the document and a copy seen by Bloomberg News.

The company said the Alibaba campaign resembled past efforts by other Chinese developers that Anthropic flagged in a blog post earlier this year.

Alibaba’s American depositary receipts sank to a session low on the news, falling more than 3% to $99.10 at 3:38 p.m. in New York on Wednesday.

Anthropic warned that Alibaba and other Chinese labs are making systematic and unauthorized use of results from leading US models to develop a rival generation of chatbots at a fraction of the cost via a practice known as adversarial distillation. It cautioned that AI systems built using this method often lack safety guardrails, and the firm urged the Trump administration to step up efforts to halt the practice.

Also Read: How Much Does a Stronger Rupee Really Reduce India’s Oil Import Bill?

“These distillation attacks are carried out illicitly, systematically, and at industrial scale to harvest US Al capabilities across frontier labs and repackage them as their own without incurring the training and research and development (R&D) costs required to train US frontier models,” Anthropic wrote in its letter.

An Anthropic spokesperson declined to enter into specifics on the letter but emphasized the importance of combating distillation through “coordinated action between government and industry.”

Anthropic’s letter marked the latest call from top American AI companies to rein in some kinds of distillation, where developers train systems using results from another AI model to create similar capabilities in a new one at a far lower cost. While tolerated for training smaller, less-advanced systems, distillation violates AI labs’ terms of use when it’s employed to replicate a cutting-edge AI model without permission.

The practice has led to to a cautious stance by US developers to the point that Anthropic, OpenAI and Alphabet Inc.’s Google have collaborated to share information about distillation attempts that violate their terms of service. Anthropic and OpenAI have each warned that Chinese AI startups, including DeepSeek and Minimax, have employed distillation to develop their own models.

Lawmakers in Washington are moving to address the US industry’s concerns. In the Senate, Tennessee Republican Bill Hagerty and New Jersey Democrat Andy Kim plan to introduce an amendment to must-pass defense legislation as soon as Wednesday that would blacklist or sanction any Chinese firm found to be improperly accessing US AI model output to help train competing models, according to a person familiar with the matter.

However, it remains unclear whether the amendment would win requisite support to be included in the defense bill’s final version. A related bipartisan bill in the House, backed by Michigan Republican Bill Huizenga and Democrat Sydney Kamlager-Dove, is also set to be considered for inclusion in the annual defense measure.

Those proposals follow initial steps by the Trump administration on the issue. In April, White House Office of Science and Technology Policy Director Michael Kratsios published a memo indicating the US would help to crack down on attempts by Chinese companies to exploit outputs from US models.