Anthropic just dropped a bombshell on Capitol Hill, accusing Alibaba of orchestrating the largest known distillation attack against its AI models and calling on Congress to tighten export controls in response.
In a letter dated June 10 sent to Senators Tim Scott and Elizabeth Warren on the US Senate Banking Committee, Anthropic detailed what it described as a systematic effort by Alibaba’s AI lab to siphon intelligence from its models. The scale is staggering: 28.8 million exchanges conducted between April 22 and June 5, routed through approximately 25,000 fraudulent accounts.
What distillation actually means
Think of AI model distillation like photocopying someone else’s homework, but at an industrial scale. A smaller, cheaper model is trained by feeding it the outputs of a more powerful one, effectively transferring the larger model’s capabilities without doing the expensive research and development work.
For Anthropic, this isn’t just an intellectual property problem. It’s a national security one. The company argues that distillation campaigns like Alibaba’s directly undermine US restrictions on advanced AI chips. If Chinese firms can simply extract the knowledge baked into American models, export controls on the hardware used to train those models become far less effective.
