Anthropic vs. Alibaba: Allegation of AI Copy Attack on Claude

According to the letter, the attackers used a method known as "distillation," in which developers of new software use artificial intelligence to train it from an already existing model. Through this process, they wanted to learn more about central aspects such as the software's architecture and its logic in executing tasks. Such access is generally prohibited by terms of use.

Demands on the US Congress

According to the letter, attackers affiliated with Alibaba interacted with Anthropic's AI software Claude nearly 29 million times through almost 25,000 fraudulent accounts from late April to early June. The concern is that AI could be used for cyberattacks. Anthropic stated that only smaller models had been affected; the core systems had not been compromised. Alibaba did not initially comment on the allegations.

Anthropic countered that the problems were not serious. At the same time, Anthropic called on Congress for, among other things, a tougher approach against Chinese companies that resort to "distillation." American AI firms should also be given the freedom to warn each other about such attacks.

Anthropic's Own Conflict with Washington

The allegations are not new. Anthropic has for some time been accusing Chinese companies of improperly acquiring knowledge from Western AI models through distillation. With the current letter, the company is now stepping up its political stance and simultaneously putting pressure on US lawmakers.

In parallel, Anthropic itself is in a conflict with the US government. The government recently forced the company to temporarily take its top artificial intelligence software off the market. The US government justified the move by citing alleged security vulnerabilities through which restrictions on the AI program called "Fable 5" could have been bypassed. Among other things, the program's ability to search for security vulnerabilities in other software is blocked. Negotiations with the US government over the block are still ongoing.

Nvidia and the Dispute Over China Exports

Anthropic is one of the sharpest rivals of OpenAI, the developer of ChatGPT, and is among the leading AI companies in the United States. In the current letter, the AI firm also reaffirms its confrontational course toward chip heavyweight Nvidia. That company is lobbying the US government to allow the export of more powerful AI systems to China. Anthropic, by contrast, demands that access by Chinese firms to technologies such as AI chips be further restricted.

Industry heavyweights outside Anthropic also view distillation critically. Tech billionaire Elon Musk recently admitted in court proceedings that he had himself used the distillation technique to train his AI chatbot Grok with models from OpenAI. The case thereby also triggered a broader debate within the US tech industry about the limits of permissible model training.

Distillation Also Controversial in the US

The allegations against Alibaba come amid a phase of growing political tensions between Washington and Beijing in the field of artificial intelligence. On both sides, accusations of unauthorized technology transfers are in the air. The US government has for some time been imposing export restrictions on high-performance AI chips to China, while Chinese firms are increasingly developing their own models that are not based on Western hardware.

With its letter to the senators, Anthropic is now trying to influence this political line. The company demands not only stricter rules against Chinese firms, but also legal assurance that American AI companies may warn each other about attacks such as the alleged distillation attack. This would enable an exchange of information that is currently restricted by competition law and antitrust regulations.

Political Dimension and Outlook

Observers assess the move as part of a broader strategy in which US AI firms want to secure their market position through political demands. Anthropic benefits from its image as a security-oriented company that pointed early to the risks of artificial intelligence. At the same time, the close interlocking of commercial interests and security-policy rhetoric raises questions.

Alibaba is one of the largest cloud and AI providers in China and is investing billions in its own AI models, which are gaining increasing international importance. Should the allegations against the company harden, this could further complicate the ongoing negotiations over export licenses for AI chips. Alibaba itself has so far not commented substantively on the allegations.

It remains unclear what concrete legal steps Anthropic intends to take against Alibaba. It is also open whether the US government will respond to the demands in the letter. What is certain is that the dispute lifts the competition between US and Chinese AI firms to a new political level – away from technical standards toward a debate shaped by security and geopolitics.

Anthropic's letter therefore marks not only another conflict with a Chinese competitor, but also the attempt to capitalize politically in Washington on the episode. Whether the demands for more cooperation among US firms and stricter rules against China actually become law now depends on Congress and on the further course of the ongoing negotiations with the US government over "Fable 5."