US agencies accuse six Chinese AI companies of industrial-scale model copying

Washington escalates an AI dispute

US law-enforcement and intelligence officials have accused six Chinese artificial-intelligence companies of systematically using outputs from American frontier models to accelerate development of their own systems. Reuters identified DeepSeek, Moonshot AI and Alibaba among the named firms, alongside MiniMax, StepFun and Z.ai. The agencies characterised the activity as aggressive, targeted and conducted at industrial scale. These are government allegations, not findings established by a court or an independently completed technical audit.

The disputed technique is model distillation, in which a smaller model learns from responses generated by a larger system. Distillation is a widely used research and engineering method, but providers can restrict automated extraction through licensing terms and technical controls. The US statement alleges that the Chinese companies used variants of models made by Anthropic, OpenAI, Google and xAI without authorisation, reducing training costs and transferring capabilities that Washington regards as strategically sensitive.

National security and intellectual property converge

US officials went beyond a commercial-intellectual-property complaint. They argued that copied capabilities could strengthen Chinese military and cyber operations and said the activity probably occurred with Beijing's awareness. Reuters has separately reported that Chinese military researchers used outputs from leading US models to train domestic systems. The latest accusation therefore connects access controls on model outputs with the wider US effort to restrict transfers of advanced chips, computing capacity and dual-use knowledge.

China rejected the characterisation. A Foreign Ministry spokesperson said the country's AI gains resulted from domestic scientific and technological development and urged Washington to stop making unsupported accusations. None of that resolves the central technical questions: which accounts generated the training data, whether contractual restrictions were breached, how much capability was transferred, and whether the activity can be attributed to the Chinese state rather than individual companies.

Diplomatic timing raises the stakes

The allegations arrived before a planned visit by Chinese President Xi Jinping to the United States later in September and a prospective bilateral dialogue on AI safety. That timing gives the dispute relevance beyond individual companies. It could influence negotiations over access to models, export controls, cyber conduct and mechanisms for reporting dangerous capabilities. It may also encourage US model providers to impose tighter rate limits, identity checks and monitoring on overseas users.

The White House's 2026 economic report describes AI as a central source of economic and strategic power and documents the large gap between US and Chinese private investment, while also noting substantial Chinese public support. That official context helps explain why Washington treats model extraction as a state-security issue. What to watch next is whether agencies release reproducible technical evidence, whether any companies face sanctions or access restrictions, and whether the dispute narrows or derails the planned safety dialogue.