Anthropic Alleges Alibaba Conducted Large-Scale Campaign to Extract AI Capabilities

Anthropic has sent a letter to U.S. senators and White House officials alleging that Alibaba orchestrated one of the largest known efforts by a Chinese company to extract capabilities from leading American artificial intelligence models. According to the letter, operators associated with Alibaba’s Qwen AI lab used nearly 25,000 fraudulent accounts to engage in over 28 million interactions with Anthropic's Claude model between April and June. This activity was focused on advanced functions such as software engineering and agentic reasoning, which Anthropic claims were targeted for replication at a lower cost through a process called adversarial distillation.

Anthropic asserts that this campaign represents the most significant effort yet by a Chinese firm to leverage U.S. AI model outputs to speed up its own development. The company warns that such practices are being carried out on an industrial scale, often resulting in systems that lack the stringent safety measures found in frontier American models. Anthropic’s letter highlights concerns over the illicit and systematic harvesting of U.S. AI capabilities by Chinese entities without bearing the substantial training and research costs required to develop these technologies domestically.

The allegations come as U.S. policymakers are considering new measures to restrict Chinese access to American AI advancements, including potential blacklisting or sanctioning of firms involved in unauthorized exploitation of U.S. models. Anthropic is urging stronger governmental action against such practices, advocating for clearer antitrust guidelines that would facilitate information sharing among U.S. companies and impose penalties on entities engaged in systematic distillation activities.