The Chinese tech giant allegedly siphoned off Claude’s capabilities through nearly 25,000 fake accounts.

Anthropic is accusing Alibaba of pulling off the largest known theft of its AI model’s capabilities to date. The claim comes from a letter sent to top U.S. senators, obtained by Reuters.

What happened

Between April 22 and June 5, 2026, Anthropic says operators tied to Alibaba and its Qwen AI lab ran a “distillation” campaign, essentially training a cheaper, weaker model by mining outputs from Claude. The numbers are staggering: roughly 28.8 million exchanges across almost 25,000 fraudulent accounts. Anthropic argues this kind of extraction helps China close the gap toward matching its top-tier Mythos Preview model. Alibaba hasn’t responded to requests for comment.

Market reaction

No major market moves yet, but the timing is sensitive. Alibaba (9988.HK) was already placed on the Pentagon’s list of Chinese military-linked companies this month – a designation it’s fighting. Any escalation in the AI feud could pressure Chinese tech stocks and add fresh volatility to U.S.-China trade sentiment.

Why it matters for traders

This isn’t an isolated spat. Anthropic flagged similar extraction attempts earlier this year from DeepSeek, Moonshot AI, and MiniMax (0100.HK), with DeepSeek’s campaign alone topping 150,000 exchanges. Washington’s response has been inconsistent – DeepSeek still isn’t blacklisted despite being flagged as a national security risk, while Commerce just restricted Anthropic’s own Mythos and Fable models over military-use fears. That regulatory whiplash is exactly the kind of uncertainty markets hate.

What to watch next

Keep an eye on the Senate Banking Committee hearing and any Commerce Department follow-up. A harder U.S. stance on Chinese AI firms could ripple through tech equities and the yuan.

Stay ahead of every market-moving headline with QuoMarkets.

Source: Reuters