Anthropic Accuses Alibaba of 28.8 Million Queries to Clone Claude

In a June 10 letter to senators, Anthropic accused operators tied to Alibaba of running 28.8 million queries across roughly 25,000 fake accounts to copy Claude's capabilities — the largest distillation campaign it says it has ever measured.

Anthropic Accuses Alibaba of 28.8 Million Queries to Clone Claude
Anthropic accused Alibaba-linked operators of distilling Claude. Image: TechRadar

Key Points

  • Anthropic told senators that Alibaba-linked operators ran 28.8 million Claude queries.
  • The alleged campaign used roughly 25,000 fraudulent accounts from April 22 to June 5.
  • It targeted Claude's agentic reasoning, software engineering, and long-horizon tasks.
  • Disclosed in a June 10 letter to Senators Tim Scott and Elizabeth Warren.

Anthropic says it caught the largest attempt yet to clone Claude. In a June 10 letter to Senators Tim Scott and Elizabeth Warren, sent a day before a Senate hearing on AI, the company shared what it called "new, confidential evidence of the largest campaign to illicitly extract Claude's capabilities we have ever measured" — and pointed at operators it says are affiliated with Alibaba and its AI lab.

The numbers Anthropic gave are large. The alleged operation ran from April 22 to June 5 and generated more than 28.8 million interactions with Claude through roughly 25,000 fraudulent accounts, a practice known as distillation: querying a rival model at scale to extract its outputs and train a competitor on them. Anthropic says the campaign zeroed in on its most valuable capabilities — agentic reasoning, software engineering, and long-horizon tasks, the exact areas where Claude leads.

This is a sharper version of a fight Anthropic has been picking all year. In February it named three Chinese labs — DeepSeek, Moonshot AI, and MiniMax — as having collectively run more than 16 million Claude interactions through about 24,000 fake accounts. The alleged Alibaba campaign, which Anthropic says is the work of the company behind the Qwen model line, dwarfs that combined total in six weeks. The timing is the tell: China is racing to close the agent-capability gap just as Washington has restricted Anthropic's most powerful systems from foreign markets, having forced the company to disable Mythos abroad under an export-control order. Anthropic is now arguing, to the lawmakers who write those rules, that the model it just demonstrated finding 10,000-plus software vulnerabilities is exactly the asset adversaries are spending millions of queries to copy. Alibaba has not publicly responded to the allegations.

Source: TechRadar, PYMNTS

Comments

Get tomorrow's roundup. Free.

One email each morning. Sneakers, sports, culture, tech.

Link copied