Anthropic, the US artificial intelligence research company, has accused the Chinese technology and e‑commerce giant Alibaba and actors connected to Alibaba’s AI laboratory, Alibaba Qwen, of attempting to extract and replicate the capabilities of its Claude language model without permission. The allegations are detailed in documents shared with Reuters.
What Anthropic says happened
Anthropic described the operation as a "knowledge distillation" experiment — a technique that speeds development of a smaller model by training it on outputs from a more advanced system. According to the company, the campaign ran from April 22 to June 5, 2026, during which roughly 25,000 fictitious user accounts were used to initiate more than 28.8 million interactions with Claude.
The company warned that such knowledge distillation could enable China to reach Anthropic’s more advanced, Mythos Preview–level technical capabilities much faster.
Previous similar attempts
Anthropic has previously reported other unauthorized extraction attempts by Chinese actors. The firm says the DeepSeek startup attempted more than 150,000 interactions in January 2025, Moonshot AI performed over 3.4 million interactions, and MiniMax conducted more than 13 million.
US response and timing
Anthropic dated the letter notifying US senators Tim Scott and Elizabeth Warren on June 10, ahead of a planned congressional hearing. In April, the White House had publicly accused China of effectively stealing intellectual property from US AI labs at industrial scale.
Two days after Anthropic’s June 10 letter, on June 12, the US Commerce Department imposed strict restrictions on Anthropic’s newest models, Mythos and Fable. The department said the move was prompted by concerns that the systems could be repurposed by Chinese or other risky foreign military and intelligence services. Following the restrictions, Anthropic blocked global access to the affected models.
Related actions and disputes
This month the US Department of Defense (the Pentagon) added Alibaba to a list of Chinese firms it says have military connections; Alibaba has strongly disputed that designation. The Commerce Department has not yet placed DeepSeek on the official trade blacklist, although a government advisory body has judged the startup a national security risk.
Why this matters
The case highlights tensions between protecting intellectual property and addressing national security risks as AI capabilities advance rapidly. Anthropic’s reported scale — over 28.8 million interactions within a limited period — represents one of the largest alleged attempts to acquire AI-related intellectual property, and it has prompted US authorities to apply tighter export and access controls to sensitive models.
Alibaba had not responded to requests for comment at the time the Reuters summary was published.
Sources
This article is based on documents and data reported by Reuters and on information released by Anthropic.
