Anthropic Letter to Senators Describes 29 Million Queries from Accounts Linked to Alibaba

Anthropic sent a letter dated June 10 to U.S. Senators Tim Scott and Elizabeth Warren stating that accounts it linked to Alibaba created thousands of fraudulent accounts and conducted nearly 29 million exchanges with its Claude AI model between April and June.

The letter described the exchanges as targeting the model's ability to handle longer and more complex tasks as well as its decision-making processes.

Anthropic stated in the letter that the activity constituted distillation attacks in which answers from a stronger model are used to train a weaker one. The letter said the campaign was conducted on an industrial scale and aimed at harvesting U.S. AI capabilities for use by Chinese firms.

Anthropic stated that distillation attacks convert hundreds of billions of dollars in American investment into a subsidy for geopolitical competitors and called the campaign the largest extraction campaign of its kind.

The letter referenced U.S. Department of Defense statements that Alibaba and several other Chinese companies are tied to the Chinese military. Alibaba has denied any links to the Chinese military. Alibaba sued the U.S. government this week to have its name removed from a Pentagon blacklist.

Anthropic urged Congress to penalize companies behind such attacks and to increase measures preventing theft of U.S. AI technology.