Anthropic Alleges Large Scale AI Theft

San Francisco-based AI developer Anthropic informed United States senators on June 10 that Alibaba-linked operators illicitly extracted proprietary capabilities from its Claude model. The company reported that these actors utilized thousands of fraudulent accounts to conduct nearly 29 million exchanges, characterizing the operation as the largest distillation attack campaign recorded to date.

The mechanism involves distillation attacks, where operators query a sophisticated model to generate data used to train and refine a secondary, less capable system. Anthropic stated that this process allows Chinese firms to harvest American research and development at a fraction of the original cost, effectively subsidizing foreign competitors with US-funded technological advancements.

Anthropic has requested that Congress implement stricter penalties and defensive measures to protect domestic AI intellectual property. The company further alleged that these activities pose broader national security risks, citing existing Pentagon designations linking major Chinese technology firms to military interests. Alibaba, which is currently challenging its inclusion on US government blacklists, has not provided a formal response to these specific allegations.