Anthropic, a US artificial intelligence company, accused Chinese tech giant Alibaba of illegally extracting advanced capabilities from its Claude AI model through a large-scale "distillation attack" conducted between April 22 and June 5, 2026 [1, 2, 3]. The attack used approximately 25,000 fraudulent accounts that interacted with Claude around 28.8 million times, according to Anthropic [2, 3, 4].
Distillation attacks work by using outputs from a stronger, high-cost AI model like Claude to train a weaker, cheaper AI model. This allows the attacker to replicate advanced capabilities such as software engineering, agentic reasoning, decision-making, and handling complex long-range tasks at lower cost [1, 5]. Anthropic stated in a letter to U.S. Senators Tim Scott and Elizabeth Warren dated June 10 that the attack specifically targeted these advanced capabilities in Claude [3, 6].
Anthropic links the operators behind the attack to Alibaba’s AI lab, known as Alibaba Qwen or 通义千问 [2, 3, 5]. The company urged the U.S. government and Congress to take legal and regulatory action against such unauthorized AI model extraction to protect American AI intellectual property and national security. An Anthropic spokesperson said, "打擊非法數據蒸餾威脅，需要政府與業界協同行動，以維護美國在人工智慧領域的領先地位" (combating illegal data distillation threats requires government and industry joint action to maintain U.S. AI leadership) [6].
Alibaba denied wrongdoing but has not publicly addressed these specific allegations. The company is currently suing the U.S. government to remove itself from the Pentagon's blacklist [1, 4].
Anthropic has previously detected similar distillation attacks by Chinese AI startups including DeepSeek, Moonshot (月之暗面), and MiniMax, which have interacted with Claude AI millions of times in attempts to extract its capabilities [3, 4]. These attacks raise concerns about subsidizing geopolitical competitors with billions of dollars of American AI research and development investments. Anthropic stated, "Distillation attacks turn hundreds of billions of dollars in American investment and [research and development] into a massive subsidy for our geopolitical competitors" [1].
Meanwhile, the U.S. Commerce Department imposed restrictions on Anthropic's latest AI models Mythos and Fable on June 12 over fears of their misuse by Chinese or other adversary military intelligence [3, 4].
Alibaba's American Depositary Receipt (ADR) share price dropped 2.73% on June 24 amid the controversy, hitting near 15-month lows [7, 8].
U.S. AI companies including OpenAI and Google are working with Anthropic through the Frontier Model Forum to detect and counter these distillation attacks [2, 3].
Anthropic's letter to U.S. Senators ahead of a June 2026 Senate Banking Committee hearing on AI detailed these allegations and called for stronger legislative measures against unauthorized AI extraction [1, 2].
