The Artificial Intelligence war is heating up.. Anthropic accuses Alibaba of exploiting Claude | technology

The American company “Anthropic” accused its Chinese counterpart, “Alibaba”, of exploiting its artificial intelligence model, “Cloud,” and accessing it illegally, then using distillation and inference techniques to train Alibaba’s artificial intelligence models, according to a report by the British newspaper “Financial Times” on June 24.

The American company described what happened as the largest distillation attack ever carried out on its services, in the accusatory letter that Anthropic sent to the US Congress.

She added that Alibaba used more than 25,000 fake accounts to generate about 28.8 million messages or interactions with its artificial intelligence tools, in a move that represents a direct violation of the company’s terms of service.

Anthropic explained that the attack, which lasted between April 22 and June 5, was aimed at accessing the capabilities of the company’s Mythos model, before the US government banned access to the model, according to a Reuters report on June 24.

The company directly accused Alibaba and its artificial intelligence laboratory, Alibaba Qwen, demanding that the Senate take the necessary measures to prevent such attacks and punish those responsible for them.

In the letter, it also expressed its support for the US government’s efforts to combat such attacks, including partnerships with artificial intelligence companies in the private sector.

This is not the first time that Anthropic has accused Chinese companies of using distillation and inference attacks to train their artificial intelligence models through Cloud and the company’s various models.

Last February, it made a similar accusation to the emerging artificial intelligence company DeepSeek, along with other artificial intelligence testers, and explained that the DeepSeek campaign included more than 150,000 exchanges, Moonshot AI included 3.4 million exchanges, and MiniMax broke the barrier of 13 million exchanges, according to a Reuters report.

The “Alibaba” group was placed on the list of companies banned by the US government, due to the company’s connection to the Chinese army and its provision of support for Chinese military activities against unspecified American targets, according to a “Financial Times” report.

For its part, Alibaba denied any connection with the Chinese Liberation Army, but did not respond to Reuters’ request to comment on Anthropic’s accusations against it.

What are distillation attacks?

A report by the American Forbes magazine describes distillation attacks as the latest weapon in the artificial intelligence war between the United States and China, because it allows those who use it to exploit the advanced capabilities of any artificial intelligence model in an effort to train and develop their own artificial intelligence model.

What happens in distillation attacks is that the attacking company collects questions and answers from the victim’s AI model, then feeds its AI models these pairs and uses them to train its models.

This method saves artificial intelligence companies the need to use large volumes of data and purchase it from various sources, as well as providing the computing and processing power needed for the process of training artificial intelligence models.

This allows AI companies to provide powerful models that exploit the capabilities of leading models, but at a much lower cost, because they do not need the initial cost of training the replica.