Anthropic accuses Alibaba of running the largest known distillation attack on Claude — 28.8 million exchanges via 25,000 fake accounts

Anthropic has accused Alibaba of orchestrating the largest known distillation attack ever conducted against its Claude AI models, disclosing the campaign in a letter sent to US Senators Tim Scott and Elizabeth Warren, as well as White House officials. The company says operators linked to Alibaba's Qwen AI lab used nearly 25,000 fraudulent accounts to conduct approximately 28.8 million exchanges with Claude between April and June 2026 — specifically targeting the model's software engineering and agentic reasoning capabilities.

The scale dwarfs every previously disclosed Chinese distillation campaign. Anthropic says the Alibaba operation surpassed the combined distillation activity previously attributed to DeepSeek, MiniMax, and Moonshot AI. Alibaba's American depositary receipts fell on the news. Alibaba has not commented on the allegations.

What distillation attacks actually do

Adversarial distillation — sometimes called model distillation or knowledge extraction — exploits the fact that AI model outputs contain implicit information about the model's internal reasoning and capabilities. By querying a target model at massive scale and training a new model on those outputs, an attacker can transfer significant capabilities from a frontier model into a cheaper, locally-controlled system — often without the safety guardrails the original model was trained to enforce.

Anthropic's letter frames this as a systematic practice, not an isolated incident: Chinese AI labs are, in its telling, using outputs from US frontier models as a low-cost shortcut to build competitive chatbots, while bypassing the safety and alignment work embedded in those models. The company explicitly warned that AI systems built this way "often lack safety guardrails."

Political and legal escalation

The disclosure arrives at a charged moment. The White House had already identified adversarial distillation as a national security concern in April 2026 — and Anthropic's letter notes the Alibaba campaign continued after that warning was issued. Senators Bill Hagerty and Andy Kim are reportedly planning legislation to sanction Chinese firms found to be improperly accessing US AI model outputs.

Alibaba is also fighting a separate battle in Washington: the company was added to the Pentagon's list of businesses allegedly supporting China's armed forces in early June 2026, a designation Alibaba has denied and is contesting in court. The distillation accusations deepen the company's US political exposure considerably.

Anthropic, backed by Google and Amazon, has positioned itself as a safety-first AI lab and has maintained a policy of restricting Claude access in China. The alleged campaign — if confirmed — represents an attempt to circumvent that restriction at an industrial scale, as first reported by Bloomberg and The Next Web on June 25, 2026.