Overview
- Anthropic told U.S. lawmakers that operators tied to Alibaba and its Qwen lab used roughly 25,000 fraudulent accounts to carry out about 28.8 million exchanges with its Claude models between late April and early June.
- The company says those queries were used for “distillation,” a technique where a smaller model is trained on a larger model’s outputs so a rival can reproduce behaviour without access to the original code or training data.
- Major U.S. AI firms began sharing threat intelligence in April after earlier distillation incidents were reported, and a White House memo described the campaigns as industrial‑scale operations by Chinese companies.
- U.S. Commerce Department actions restricting foreign access to some Anthropic models led the company to disable affected models globally and have prompted legal and diplomatic pushback while technical negotiations continue.
- Alibaba has not publicly responded and independent verification of Anthropic’s specific accusation is not yet available, leaving attribution unresolved and raising questions about how companies will balance model access with defensive testing.