Anthropic, the AI safety company behind the Claude model, has accused Chinese tech giant Alibaba of orchestrating a large-scale distillation attack. The accusation was detailed in a letter sent to U.S. senators earlier this month, according to an Inc report.
Model distillation is a process where a smaller model is trained to replicate the behavior of a larger, more powerful one — often by querying it extensively and using the outputs as training data. In this case, Anthropic claims Alibaba systematically extracted knowledge from its models without authorization.
The attack raises concerns about intellectual property theft and the security of frontier AI systems. Distillation can effectively steal months of expensive research and development, allowing competitors to build similar capabilities at a fraction of the cost.
This incident highlights growing tensions in the AI arms race between U.S. and Chinese companies. As models become more powerful, the economic and strategic incentives for copying them through distillation will only intensify.
Alibaba has not publicly responded to the accusation. The incident underscores the need for stronger technical and regulatory protections against model theft as AI capabilities advance.