Anthropic Accuses Alibaba of Mass-Distilling Claude's AI
Anthropic told US officials that Alibaba used ~25K accounts to hit Claude 28.8M times in three months for adversarial distillation.
Anthropic has gone to the US government with a serious allegation: Alibaba ran a massive adversarial distillation campaign against Claude.
In a letter to US officials, Anthropic accused the Chinese tech giant of accessing Claude approximately 28.8 million times between April and June. The operation allegedly used around 25,000 accounts to carry out what Anthropic described as an illicit large-scale effort.
Adversarial distillation is a technique where one AI model's capabilities are systematically extracted by querying it at scale — essentially training a competing model on another's outputs.
The sheer volume is staggering. Nearly 29 million API hits across roughly three months suggests an industrial-level operation, not casual usage. That Anthropic escalated this directly to government officials signals how seriously the company views the threat.
No word yet on any official response from Alibaba or US authorities.