It is the largest distillation attack ever recorded

Anthropic has sent a confidential letter to the US Senate accusing Alibaba of orchestrating the largest distillation attack it has ever suffered Claude. According to share Bloomberg, the Chinese company would have created almost 25,000 fake accounts to extract capabilities from its AI model without paying for them or investing in its own development.

What exactly happened. The media, which had access to the letter, affirms that this dates back to June 10 and was addressed to Tim Scott and Elizabeth Warren. It states that operators linked to Alibaba and its AI laboratory, Alibaba Qwenwould have generated more than 28.8 million interactions with Claude between April 22 and June 5. According to the media, to avoid detection, they resorted to obfuscation techniques and proxy networks.

The objective of these attacks, according to the media, was to extract the most advanced capabilities of the model, including agentic reasoning and software engineering. Claude is not available in China, so these attacks were supposedly executed through fraudulent accounts.

What is distillation about?. Is a widely used technique in the industry to train smaller and cheaper models from the results of a more powerful one. In itself it is not illegal, but it is when it is done without authorization and on a massive scale as would have happened, in theory, in this case. Anthropic argues that these attacks turn “hundreds of billions of dollars in American investment and development into a massive subsidy for our geopolitical competitors,” according to collect the Wall Street Journal.

It’s not the first time. In February, Anthropic had already pointed out three Chinese laboratories (DeepSeek, Moonshot and MiniMax) for a similar campaign that claims to have generated more than 16 million interactions with their model through some 24,000 fake accounts. The signature account that OpenAI and Google have also published findings on similar attacks against their own models.

What makes the Alibaba case different is the scale, almost twice as many interactions as the previous case. Also the timing, since the information suggests that it occurred weeks after the Trump administration would warn explicitly in an internal memo that this type of AI theft and at this scale was “unacceptable.”

Audacity. Anthropic accuses Alibaba of having acted “brazenly” knowing it was in the crosshairs. The Chinese company is listed on the New York Stock Exchange, operates in the United States and has investors regulated by the SEC. Even so, the media suggest that the attack occurred after the White House statement.

The Financial Times points out Furthermore, the White House had concluded that Alibaba provided technological support for Chinese military operations against US targets, something the company categorically denies. This week, Alibaba He also filed a lawsuit against the Pentagon to remove her from a blacklist that links her to the Chinese army.

What Anthropic asks of Congress. The company proposes three concrete measures:

• The first is to update antitrust laws so that AI companies can share information about these types of tactics without incurring legal problems.
• The second, strengthen export controls on advanced chips so that China cannot take advantage of the answers provided by American AI models even if it obtains them.
• The third, legislate to directly penalize Chinese laboratories that carry out distillation attacks, with sanctions that could include blocking their access to AI chips or models manufactured in the United States.

The race behind. Anthropic warns in its letter that if China manages to close the gap, it could have “advanced cyber capabilities to deploy against the US government and American companies.” The model Mythos from Anthropic, which has worried everyone with its capabilities, it is precisely the one that worries Beijing the most.

360 Security Technology founder Zhou Hongyi called Mythos a “cyber nuclear weapon” at a conference in Beijing, according to informed South China Morning Post. Zhou admits that China is “far below” that level and called for developing its own equivalent to ensure mutually assured destruction in the event of a digital conflict.

What can happen now. Alibaba shares have fallen more than 4% in Hong Kong after the accusation was made public, according to they advanced From the WSJ Laila Khawaja, director of research at Gavekal Technologies, told the media that, given that this is not the first accusation of this type against Chinese companies, the reputational damage should be limited. What’s really at stake is whether Anthropic can convince Congress to greenlight its requests. For now, the company simply says that it “will continue to work with Congress and the Administration to maintain American leadership in AI.”

Cover image | AI-generated with Magnific

In Xataka | The great fortunes of artificial intelligence: who is who in the new AI millionaires