The Trump administration has released an AI refinement crackdown plan, accusing Chinese companies of systematically stealing model capabilities.

On April 23, the White House Office of Science and Technology Policy (OSTP) Assistant to the President Michael J. Kratsios released an official statement saying the Trump administration has information showing that foreign entities (mainly located in China) are deliberately targeting major U.S. artificial intelligence companies by systematically extracting U.S. AI model capabilities through “tens of thousands of proxy accounts” and jailbreak technology systems, while also announcing four response measures.

Attack Methods and Security Warnings Confirmed by the White House OSTP

According to the White House OSTP official statement, the attack methods used by foreign entities include two tactics: using “tens of thousands of proxy accounts” to evade detection, and “leveraging jailbreak technology to leak proprietary information.” The OSTP statement’s original wording says: “These coordinated actions systematically extract capabilities from U.S. artificial intelligence models, using U.S. expertise and innovation.”

In his statement, Kratsios said that models developed through such distillation activities “cannot completely replicate the performance of the original version,” but they “do enable foreign vendors to roll out products that appear quite competitive on certain benchmarks at very low cost.” The White House also warned that models developed through distillation may lack safety protocols and no longer be “neutral and committed to truth.”

Anthropic Accuses Three Chinese AI Companies: DeepSeek, Moonshot, MiniMax

According to an official statement released by Anthropic in February 2026, the company accuses Chinese AI companies DeepSeek, Moonshot, and MiniMax of launching distillation attacks against its Claude model using approximately 24,000 scam accounts, generating more than 16 million transaction request attempts. According to Anthropic’s official announcement, the capabilities allegedly extracted include:

· Intelligent reasoning capabilities

· Code generation and data analysis

· Rating tasks based on scoring criteria

· Computer vision capabilities

The Trump Administration Announces Four Response Measures

According to the White House OSTP official statement, the Trump administration plans to respond to foreign AI distillation threats through the following four measures: sharing with U.S. AI companies information related to potential large-scale attacks; helping the private sector build more effective coordinated response mechanisms; partnering with the private sector to establish strong technical defenses; and exploring related measures to “hold foreign actors accountable.” The statement did not disclose specific legal action timelines or targets for sanctions.

Frequently Asked Questions

What specific methods are used in the “AI distillation attack” referred to in the White House OSTP statement?

According to the White House OSTP official statement, the attack methods include two categories: using “tens of thousands of proxy accounts” to evade detection mechanisms used by U.S. AI companies, and using jailbreak technology to bypass model safety restrictions to leak proprietary information, with the goal of systematically extracting core capabilities of U.S. AI models.

What are the three Chinese companies accused by Anthropic, and what are the allegations?

According to an official statement released by Anthropic in February 2026, the three accused companies are DeepSeek, Moonshot, and MiniMax. They are alleged to have launched distillation attacks against the Claude model using approximately 24,000 scam accounts, generating more than 16 million transaction request attempts. The capabilities allegedly extracted include multiple areas such as intelligent reasoning, code generation, and computer vision.

Has the Trump administration taken specific legal action against the relevant Chinese companies?

According to the White House OSTP official statement, as of the date the statement was released, the government is “exploring measures to hold foreign actors accountable,” and the statement did not disclose any specific executed sanction decisions or legal prosecution actions.