What are AI distillation attacks and why are US startups worried about them?

Anthropic recently accused China's Alibaba of making 29 million exchanges with Claude using about 25,000 fraudulent accounts. The technique, known as a distillation attack, allows an adversary to train a smaller clone model on responses generated by a stronger model. US AI startups argue that such attacks allow Chinese competitors to rival American technology at a fraction of the cost.