Anthropic report accuses Chinese AI labs of distillation attacks

Anthropic published a report alleging sustained distillation attacks against its models from China-based companies including Alibaba, Moonshot AI, and DeepSeek. The report says these attacks have intensified in recent months as competition has grown.

Why it matters: Distillation, training a cheaper model to mimic a frontier model's outputs, has been central to how Chinese labs like DeepSeek have closed the capability gap cheaply; this report frames it as adversarial rather than incidental. Combined with recent news of Chinese chipmakers gaining share and DeepSeek's efficient new releases, it points to an escalating US-China contest over model IP as much as hardware.