Anthropic Says Alibaba Used 25,000 Fake Accounts to Copy Its Most Powerful AI Features

TLDR

• Anthropic accuses Alibaba of running the largest known AI “distillation attack” against its Claude model
• Nearly 29 million exchanges were made through around 25,000 fraudulent accounts between April 22 and June 5, 2026
• The attack aimed to extract Claude’s advanced capabilities to train Alibaba’s own AI models
• Anthropic sent a letter to US Senators Tim Scott and Elizabeth Warren urging tougher action
• The US Commerce Department has since restricted Anthropic’s Mythos and Fable AI models globally

Anthropic has accused Chinese tech giant Alibaba of running a large-scale campaign to steal capabilities from its Claude AI model. The San Francisco-based AI company says Alibaba used a method called “distillation attacks” to extract valuable AI outputs and use them to train weaker models.

The campaign ran between April 22 and June 5, 2026. During that time, operators linked to Alibaba and its AI lab, Alibaba Qwen, generated more than 28.8 million exchanges with Claude through nearly 25,000 fake accounts.

Anthropic described this as the largest known attack of its kind against the company.

What Is a Distillation Attack?

A distillation attack works by feeding questions to a more powerful AI model and using those answers to train a less capable one. It is a way to copy advanced AI abilities without doing the expensive research behind them.

Anthropic says the attackers targeted Claude’s most valuable features. These included its ability to handle complex, long-form tasks and its decision-making approach.

The company said these attacks are carried out on an “industrial scale” and turn hundreds of billions of dollars in American AI investment into a resource for geopolitical rivals.

Anthropic Calls for Congressional Action

Anthropic sent a letter dated June 10 to US Senators Tim Scott and Elizabeth Warren. Both sit on the Senate Banking Committee, which had a hearing on AI scheduled around that time.

In the letter, Anthropic urged Congress to penalize companies behind such attacks and strengthen protections for US AI technology.

This is not the first time Anthropic has raised the alarm. In February, the company said it had identified similar campaigns by Chinese AI startup DeepSeek, Moonshot AI, and MiniMax. The DeepSeek operation involved over 150,000 exchanges, while MiniMax’s reached over 13 million.

Alibaba did not respond to requests for comment. The company was added to the Pentagon’s list of Chinese military-linked companies this month and is challenging that designation in court.

The White House said in April that China was stealing US AI intellectual property on an industrial scale. OpenAI has also previously accused Chinese groups of using the same distillation method.

On June 12, two days after Anthropic’s letter was sent, the US Commerce Department placed restrictions on Anthropic’s Mythos and Fable AI models. Officials said they feared the models could be used by military intelligence in China and other countries. Anthropic disabled global access to both models as a result.

The post Anthropic Says Alibaba Used 25,000 Fake Accounts to Copy Its Most Powerful AI Features appeared first on CoinCentral.