Anthropic told US senators that Alibaba's Qwen lab used about 25,000 fake accounts to copy Claude's capabilities.
The alleged campaign ran 28.8 million exchanges between April 22 and June 5, using proxies to dodge China access blocks.
Anthropic calls it the largest known distillation attack on its models, but it is an accusation Alibaba has not answered.
Anthropic has accused Chinese tech giant Alibaba of running a massive operation to copy its Claude AI. The viral version calls it the largest AI theft in history. The accurate version, in Anthropic's own words, is the largest known distillation attack on the company to date. The story is real and serious, so the precise framing matters.
What Anthropic alleges
In a June 10 letter to the US Senate Banking Committee, addressed to Chair Tim Scott and Ranking Member Elizabeth Warren, Anthropic said operators affiliated with Alibaba and its Qwen AI lab made 28.8 million exchanges with Claude using roughly 25,000 fraudulent accounts.
The activity ran from April 22 to June 5, 2026. Bloomberg first reported the letter, and CNBC confirmed it on June 24. The campaign targeted Claude's most valuable skills: agentic reasoning, software engineering, and long-horizon task completion.
How the alleged attack worked
The method is called distillation. You repeatedly prompt a stronger model, collect its answers, then train a weaker model on them to copy its abilities without paying to build them.
Anthropic says Alibaba used proxy networks and obfuscation to get around the geographic restrictions that block Chinese entities from Claude, which is the part the screenshot refers to when it says Alibaba bypassed the China block. Anthropic argues these attacks turn billions in American R&D into a subsidy for competitors.
Why the framing needs a caveat
This is an accusation, not a proven case. Alibaba has not publicly responded to the distillation claim, and a representative did not reply to requests for comment. Analysts also called the market reaction muted, with Alibaba's US shares slipping about 2.7%. Anthropic is also an interested party making the case to lawmakers, and it has filed similar complaints before, naming DeepSeek, Moonshot, and MiniMax in February.
What happens next
The letter is already moving policy. Senators are weighing an amendment to penalize entities caught running distillation campaigns. The timing is awkward for Anthropic too, since the US government recently ordered it to suspend access to its most advanced Mythos and Fable models, a separate dispute it is still working to resolve.
Embeds:
YouTube (news report on the accusation):
https://www.youtube.com/watch?v=VcdLKbrdHc0
X (verified post on the story):
https://x.com/kimmonismus/status/2069879640835961277
Quick Links:
