US Warns China Is Scaling Up AI Technology Theft Through Distillation Tactics
The White House has accused China-based entities of systematically stealing US artificial intelligence technology, describing the activity as happening at industrial scale.
The allegation was outlined in a memo prepared by Michael Kratsios, director of the White House Office of Science and Technology Policy, according to the Financial Times.
The memo argues that foreign actors, primarily headquartered in China, are targeting valuable US intellectual property tied to advanced AI models. US officials say they are weighing steps to deter and punish entities involved, while helping American companies defend their systems.
Why distillation has become a flashpoint
At the center of the dispute is AI distillation, a technique used to train smaller models by learning from the outputs and capabilities of larger, more expensive systems.
In legitimate settings, distillation is widely used to cut computing costs and make models faster to run.
US officials and some AI firms say the same approach can be abused to replicate protected model behavior without permission, effectively extracting know-how embedded in leading American systems.
The concern has grown as top-tier US models have become key commercial and national security assets.
DeepSeek and industry complaints add pressure
The issue drew broader attention amid claims that the Chinese chatbot DeepSeek was trained using distillation based on advances made by US companies.
The memo referenced by the Financial Times says the US will review measures to hold foreign entities accountable for large-scale distillation campaigns.
Chris McGuire, a technology expert at the Council on Foreign Relations, told the newspaper that Chinese companies may use distillation to compensate for gaps in compute and to reproduce core capabilities of US models. He argued that the tactic can accelerate catch-up efforts while sidestepping the cost of building frontier systems from scratch.
US AI developers including OpenAI and Anthropic have increasingly complained about suspected distillation-driven scraping and automated querying aimed at their models.
Anthropic has publicly said it has tracked attempts it attributes to Chinese groups, naming firms such as DeepSeek, Moonshot AI and MiniMax in a February complaint referenced by the Financial Times.
Beyond lost revenue and eroded competitive advantage, US companies also warn of safety risks if distilled models are released without comparable safeguards. Industry voices say weaker guardrails could increase the chance that powerful systems are misused for harmful tasks, including assistance with biological threats or malicious cyber activity.
