At the center of the new conflict was distillation, which allows you to build competitors on the answers of foreign systems.
The White House has accused Chinese players of systematically trying to “surrive” key opportunities from American advanced AI models. According to Washington, we are not talking about one-time experiments, but about a large-scale and well-organized campaign that allows you to obtain valuable intellectual property without direct access to the original systems.
The statement was made at an inconvenient moment. In a few weeks, Donald Trump is expected to meet with Chinese President Xi Jinping in Beijing, and a new round of arguing around artificial intelligence risks turning into one of the most pressing topics ahead.
Details of the charges were presented by the director of the White House Science and Technology Policy Office, Michael Krazios. In a memo to the federal agencies, Kracios wrote that the U.S. government has data on the activities of foreign structures, mainly related to China, which conduct targeted industrial-scale campaigns to “distillate” American advanced AI systems.
The meaning of such a scheme is reduced to a huge number of requests for someone else's model. Attackers create proxy accounts in order not to fall under the locks, then check the behavior of the system on a variety of wording and apply methods for bypassing restrictions to pull out closed or service answers. After millions of such appeals, you can collect a large array of data and understand how the model reacts to certain commands.
Then comes distillation, one of the most sensitive techniques for the industry. A more compact model is trained on the responses of a large and expensive system, as a result of which the developer receives similar opportunities at a much lower cost. According to the White House, this approach allows the use of American achievements without spending years and a lot of money to create their own advanced models.
Washington fears not only the leakage of opportunities, but also the destruction of protective mechanisms. Officials believe that such campaigns can weaken the built-in barriers that should keep the model within safe and controlled responses. In other words, we are no longer just talking about copying the results, but about the risk of transferring the behavior of the system without the restrictions that the creators laid.
The Chinese Embassy in Washington denied the accusations. The Chinese side called the statements groundless and said that Beijing attaches great importance to the protection of intellectual property rights.
The new dispute could further exacerbate the technological confrontation between the United States and China. The issue is already going far beyond the laboratories and touches on solutions for the supply of advanced chips, including NVIDIA products. The U.S. administration has previously authorized conditional sales, but deliveries, according to officials, have not yet begun.
American AI companies have also complained of similar attempts in recent months. OpenAI and Anthropic reported large-scale attacks with signs of distillation and linked such activity to Chinese firms, including DeepSeek, Moonshot AI and MiniMax. In parallel, the United States continues criminal prosecution in cases related to the leakage of AI development. In 2024, U.S. prosecutors indicted a former Google engineer who was suspected of stealing trade secrets in the field of AI and transferring materials to Chinese companies.
At the same time, Kracios doubted that the copied models will remain effective for a long time. According to the official, as defense technologies develop, it will be easier to identify such schemes, and models collected on the basis of “squeezed” opportunities will begin to lose value faster. The White House has already stated that it intends to share intelligence with American AI companies and look for ways to bring the perpetrators to justice.
Against the background of these statements, the rivalry between the United States and China in the field of artificial intelligence is moving into an even more rigid phase. If earlier the struggle followed chips, investments and personnel, now the models themselves, their behavior and the ability of states to protect the results of the most expensive technological developments were at the center of the conflict.
The White House has accused Chinese players of systematically trying to “surrive” key opportunities from American advanced AI models. According to Washington, we are not talking about one-time experiments, but about a large-scale and well-organized campaign that allows you to obtain valuable intellectual property without direct access to the original systems.
The statement was made at an inconvenient moment. In a few weeks, Donald Trump is expected to meet with Chinese President Xi Jinping in Beijing, and a new round of arguing around artificial intelligence risks turning into one of the most pressing topics ahead.
Details of the charges were presented by the director of the White House Science and Technology Policy Office, Michael Krazios. In a memo to the federal agencies, Kracios wrote that the U.S. government has data on the activities of foreign structures, mainly related to China, which conduct targeted industrial-scale campaigns to “distillate” American advanced AI systems.
The meaning of such a scheme is reduced to a huge number of requests for someone else's model. Attackers create proxy accounts in order not to fall under the locks, then check the behavior of the system on a variety of wording and apply methods for bypassing restrictions to pull out closed or service answers. After millions of such appeals, you can collect a large array of data and understand how the model reacts to certain commands.
Then comes distillation, one of the most sensitive techniques for the industry. A more compact model is trained on the responses of a large and expensive system, as a result of which the developer receives similar opportunities at a much lower cost. According to the White House, this approach allows the use of American achievements without spending years and a lot of money to create their own advanced models.
Washington fears not only the leakage of opportunities, but also the destruction of protective mechanisms. Officials believe that such campaigns can weaken the built-in barriers that should keep the model within safe and controlled responses. In other words, we are no longer just talking about copying the results, but about the risk of transferring the behavior of the system without the restrictions that the creators laid.
The Chinese Embassy in Washington denied the accusations. The Chinese side called the statements groundless and said that Beijing attaches great importance to the protection of intellectual property rights.
The new dispute could further exacerbate the technological confrontation between the United States and China. The issue is already going far beyond the laboratories and touches on solutions for the supply of advanced chips, including NVIDIA products. The U.S. administration has previously authorized conditional sales, but deliveries, according to officials, have not yet begun.
American AI companies have also complained of similar attempts in recent months. OpenAI and Anthropic reported large-scale attacks with signs of distillation and linked such activity to Chinese firms, including DeepSeek, Moonshot AI and MiniMax. In parallel, the United States continues criminal prosecution in cases related to the leakage of AI development. In 2024, U.S. prosecutors indicted a former Google engineer who was suspected of stealing trade secrets in the field of AI and transferring materials to Chinese companies.
At the same time, Kracios doubted that the copied models will remain effective for a long time. According to the official, as defense technologies develop, it will be easier to identify such schemes, and models collected on the basis of “squeezed” opportunities will begin to lose value faster. The White House has already stated that it intends to share intelligence with American AI companies and look for ways to bring the perpetrators to justice.
Against the background of these statements, the rivalry between the United States and China in the field of artificial intelligence is moving into an even more rigid phase. If earlier the struggle followed chips, investments and personnel, now the models themselves, their behavior and the ability of states to protect the results of the most expensive technological developments were at the center of the conflict.
