A Reuters review found Chinese military-linked researchers used outputs from OpenAI and Anthropic models to train domestic AI systems. The findings sharpen a dispute over unauthorised capability extraction ahead of US-China talks on AI governance.