Anthropic Accuses Chinese Giant Alibaba of 'Distilling' Claude: One of AI's Biggest Theft Claims

Anthropic has accused Chinese e-commerce and tech giant Alibaba of copying its Claude model. According to the claim, Alibaba created a mass of fake accounts and sent Claude millions of questions, then harvested the answers to train and strengthen its own model -- a technique known as distillation, one DeepSeek reportedly used as well.

The numbers Anthropic cited aren't small: roughly 25,000 fake accounts were created in just 45 days, generating nearly 30 million interactions with Claude. The questions specifically targeted Claude's coding ability and its capacity to complete tasks autonomously.

Anthropic reported the matter directly to the US Senate in a letter. This isn't the first such episode -- back in February, DeepSeek, Moonshot, and MiniMax faced similar accusations from Anthropic, OpenAI, and Google over the same kind of technique. Sanctions rarely follow these accusations, but the relationships cool and technology trade between the parties gets constrained.

What Anthropic wants from the US government is stricter chip export controls and clearer rules. CEO Dario Amodei said at the G7 summit that China's AI rise needs to be stopped, framing it as a matter of national security.

The core question is whether China's rapid rise in AI reflects genuinely independent effort or is partly the product of copying. As restrictions tighten, more of the picture should come into focus -- though China's own government is also advancing its domestic development rapidly enough that it may eventually not need to copy at all.

The same segment also covers OpenAI and Broadcom's first jointly developed inference chip, 'Halapeno' (likely meaning Jalapeño), which went from design to completion in just nine months -- a sign that the chip race among AI companies is accelerating too.