Anthropic says Alibaba illicitly pulled from Claude

https://www.taipeitimes.com/News/biz/archives/2026/06/26/2003859739

Publish Date: 2026-06-25 12:00:00

Here are the key points from the article summarized below:

– Anthropic, a US-based AI company, has accused Alibaba Group of an illicit transfer of its AI capabilities using a technique known as “distillation,” which trains a less capable model on more powerful outputs.
– The “distillation” campaign by Alibaba and its AI lab Qwen allegedly took place between April 22 and June 5, involving over 25,000 fraudulent accounts making 28.8 million exchanges with Anthropic’s Claude AI model.
– Anthropic’s letter, dated June 10 and addressed to US senators Tim Scott and Elizabeth Warren, highlighted Alibaba’s actions as the largest known attack of this kind on the company.
– The accusation comes amidst ongoing concerns about Chinese companies extracting technology from US counterparts, a matter already raised by the White House in April regarding industrial-scale theft of intellectual property from US AI labs.
– This issue is gaining attention ahead of an AI-related hearing in the US Senate. Anthropic supports the government’s efforts to combat such attacks and is working with other AI companies and the global AI community.
– Earlier in the year, Anthropic also discovered campaigns by other Chinese AI labs (DeepSeek, Moonshot AI Technology Co, and MiniMax Group Inc) attempting to steal capabilities from its AI platform.
– The US Department of Commerce recently mandated Anthropic to restrict foreign nationals from accessing its most advanced AI platforms, underlining a shift towards more aggressive regulatory measures on AI by the US government.