Anthropic Accuses Alibaba of Illicit AI Capability Extraction
AI firm alleges 'brazen' campaign using nearly 29 million fraudulent interactions to train rival models.
US artificial intelligence company Anthropic has accused Alibaba, the Chinese e-commerce and technology giant, of "brazenly" and "illicitly" extracting the capabilities of its Claude AI model. In a letter addressed to US Senators Tim Scott and Elizabeth Warren, Anthropic stated that operators linked to Alibaba conducted nearly 29 million exchanges with Claude using thousands of fraudulent accounts. The company characterized this as the largest extraction campaign of its kind.
Anthropic detailed that the campaign utilized "distillation attacks," a method where answers from a more powerful AI model are extracted to train a less advanced one. According to the AI firm, Alibaba-linked operators specifically targeted Claude's ability to handle lengthy and complex tasks, as well as its decision-making processes. Anthropic claims these attacks are conducted on an "industrial scale" to enable Chinese companies to acquire and repurpose US AI technology.
The company also cited other alleged attacks that it asserts pose a threat to the US military. Anthropic argued that such "distillation attacks" effectively turn substantial American investment in AI research and development into a "massive subsidy for our geopolitical competitors." The letter referenced claims by the US Department of Defense that Alibaba and other major Chinese firms, including BYD and Baidu, are connected to the Chinese military. These companies have denied such allegations, and Alibaba has recently filed a lawsuit against the US government to challenge its inclusion on a Pentagon blacklist.
This is not the first time US AI developers have accused Chinese competitors of using distillation attacks to develop rival AI technologies at a significantly lower cost. OpenAI has also previously alleged that Chinese groups have employed similar practices. Anthropic, a leading AI developer alongside OpenAI, is preparing for a significant stock market debut that could establish it as one of the world's most valuable companies. The firm's more advanced models have previously raised cybersecurity concerns regarding their potential to exploit weaknesses in computer systems.