The FBI, National Security Agency and Cybersecurity and Infrastructure Security Agency accused Chinese developers of extracting capabilities from leading U.S. artificial intelligence models. Their joint advisory named DeepSeek, Alibaba, Moonshot AI and Z.ai and said the activity had occurred since at least late 2024, likely with Chinese government awareness.
U.S. agencies said developers routed distillation requests through multiple pathways and shared premium subscriptions across teams to gain access and reduce costs. China’s Commerce Ministry called the allegations groundless, described distillation as a common global practice and said Beijing would take countermeasures if companies were suppressed.
The Chinese Foreign Ministry urged cooperation and rejected what it called smears, while the named companies did not immediately respond to Associated Press requests for comment. The dispute emerged before expected talks between President Donald Trump and Chinese leader Xi Jinping in which AI governance is expected to be discussed.
Model distillation can be a legitimate training method, but providers may allege abuse when automated access violates service terms or reproduces restricted proprietary capabilities. Technical evidence would need to distinguish learning from public outputs, prohibited account behavior and the reproduction of protected systems or data. U.S. export restrictions already limit China’s access to some advanced chips, increasing the strategic significance of model efficiency and access to hosted systems.
At the edition deadline, the agencies did not publicly release all supporting logs, and China’s denial did not address each technical allegation in the advisory. The next documented developments will be any sanctions, access restrictions or indictments tied to the advisory and whether Trump-Xi discussions produce a bilateral channel or further retaliation.