US Bill Targets 'Distillation Attacks' That Copy Leading AI Models

A new bipartisan bill in the US House of Representatives would task the federal government with creating formal guidance on defending AI models against distillation attacks. The AI Cyber Defense Act, shared exclusively with CyberScoop, is sponsored by Reps. Pat Harrigan (Republican) and George Whitesides (Democrat). It would direct the National Institute of Standards and Technology (NIST) to develop voluntary industry standards and best practices for preventing such attacks from foreign entities in four countries: Russia, China, Iran and North Korea.

A distillation attack happens when one AI model is trained on the output data and reasoning of another, usually more advanced, model. The US government, along with Anthropic, OpenAI and Google, has accused Chinese companies including Moonshot AI, DeepSeek and Alibaba of running "systematic" campaigns to distill and copy US commercial models. Harrigan told CyberScoop that keeping a six-to-12 month lead is incredibly expensive for US industry, and that copying lets others keep their own spending low.

As a first step, NIST would have to define what a distillation attack is. That definition must separate it from the sharing and collaboration between open-weight and open-source models that has also driven innovation. The source excerpt describes the guidance as voluntary, so it would not set binding rules on its own.