Anthropic Accuses Alibaba of a Massive Attack on Claude

Anthropic Accuses Alibaba of Massive Attack on Claude

Anthropic has sent a letter to the U.S. Senate accusing Alibaba of orchestrating “the largest known distillation attack” against its AI models. 28.8 million exchanges, thousands of fraudulent accounts: the offensive is documented, quantified, and public.

Stealing the capabilities of an AI model without copying a single line of code: this is exactly what Anthropic accuses Alibaba of. In a letter addressed on June 10 to the U.S. Senate, the publisher of Claude accuses the Chinese giant of conducting “the largest known distillation attack to date” against its systems. The method: operators affiliated with Alibaba and its AI lab allegedly used thousands of fraudulent accounts to accumulate 28.8 million exchanges with Anthropic's models, aiming to train their own systems on these outputs. This is not an isolated incident detected internally. It is a public accusation, conveyed to lawmakers, with figures to support it.

Distillation, an Industrial Catch-Up Technique

Distillation is not new in the AI sector. The principle is simple: rather than training a model from scratch on raw data, it is trained to reproduce the responses of a more powerful model. The result is a more compact system that captures some of the expertise of the source model, without having invested in its fundamental research. It is legal when the terms of use allow it. It is not when one goes through thousands of fraudulent accounts to circumvent a competitor's safeguards.

What Anthropic describes resembles less a classic security breach than a structured operation. The terms used in the letter, “brazenly” and “illicitly,” leave little room for ambiguity: for Anthropic, this is not an accidental misuse, but a deliberate campaign. Context matters: Alibaba has been accelerating in generative AI for several quarters, and its Qwen models are progressing at a pace that has surprised more than one Western observer.

Anthropic Takes a Stand and Chooses the Political Arena

The choice of recipient is not trivial. Addressing this letter to the U.S. Senate Committee on Banking, signed by Senators Tim Scott and Elizabeth Warren, places the matter on geopolitical as well as legal ground. Anthropic is not content with blocking accounts or filing civil lawsuits: the company is alerting lawmakers, in a context where tensions between Washington and Beijing over AI technologies have never been higher. As reported by the Financial Times, the letter has been obtained and confirmed by several U.S. media outlets.

This is not the first time Anthropic has raised its voice on this issue. The company had previously implicated Chinese labs for similar practices, with mixed reactions from the tech community at the time. This time, the figures are precise, the target is named, and the chosen channel is institutional. The startup, valued at several tens of billions of dollars, understands that protecting its models cannot rely solely on technology. There must also be rules and arbiters.

Alibaba has not yet publicly responded to the accusations. But the case raises a question that the entire sector will have to confront: if large-scale distillation becomes a standard catch-up strategy, how can players investing billions in fundamental research protect their lead other than by running faster?