The United States is reportedly seeking to warn other countries about alleged efforts by China-based artificial intelligence companies to train their models with U.S. AI businesses’ intellectual property.
A State Department cable from Friday instructed U.S. diplomatic staff to tell foreign counterparts about “concerns over adversaries' extraction and distillation of U.S. A.I. models”, per Reuters. The distillation process trains AI tools by using thousands of accounts with a larger AI model to expose and save its training data.
“A separate demarche request and message has been sent to Beijing for raising with China,” according to the cable.
The State Department’s cable named DeepSeek, Moonshot, and Minimax, all Chinese businesses, claimed they had trained their models using distillation. China has denied allegations that Chinese AI companies are stealing U.S. intellectual property.
U.S.-based OpenAI has accused Chinese company DeepSeek of training its models based on OpenAI’s data. DeepSeek denied this, saying that its V3 model was trained using data it had collected naturally from the internet.
Anthropic also said in February that it had found “industrial-scale” efforts to distill its data by DeepSeek, Moonshot, and MiniMax.
The White House wrote in a separate internal memo last week that it would share more information with U.S. AI companies about the actors involved and tactics used in distillation campaigns.
The memo said that the White House would explore options to hold foreign actors accountable for their distillation efforts, though it did not describe these in detail.
