Administration Threatens Action Over AI 'Distillation'
The Trump administration has threatened to take action against Chinese artificial intelligence companies accused of systematically extracting capabilities from US AI models through a technique known as "distillation." In a memo released Thursday, Michael Kratsios, the president's chief science and technology adviser, warned that "surreptitious, unauthorised distillation campaigns" were enabling foreign entities "principally based in China" to release models matching leading US systems at a fraction of the cost (Source: BBC; Source: SCMP).
The White House announced a four-point plan to combat what it described as "industrial-scale campaigns" by foreign actors: sharing information with US AI companies about tactics and actors involved; better coordination with companies to fight attacks; developing best practices to identify, mitigate and remediate such activities; and exploring how to hold foreign actors accountable (Source: BBC). The memo did not specify what penalties might be imposed.
Diplomatic Mobilization and Congressional Scrutiny
The State Department has ordered a global diplomatic push to raise awareness of alleged Chinese IP theft, according to a cable dated Friday seen by Reuters. The instructions direct diplomatic staff worldwide to discuss "concerns over adversaries' extraction and distillation of US AI models" with foreign counterparts, with a separate demarche sent to Beijing (Source: Dawn).
On Capitol Hill, the House Select Committee on China released a report titled "Buy What It Can, Steal What It Must: China's Campaign to Acquire Frontier AI Capabilities." Committee Chair John Moolenaar alleged: "China is dependent on our tech stack to continue their AI development. China is willing to buy what they can, and steal what they cannot, to advance their AI ambitions" (Source: SCMP).
A Senate Judiciary Committee hearing titled "Stealth Stealing: China's Ongoing Theft of US Innovation" featured testimony from Helen Toner, interim executive director at Georgetown University's Centre for Security and Emerging Technology and former OpenAI board member, who stated there was "strong evidence" that Chinese AI firms are using distillation techniques to extract capabilities from US models (Source: SCMP).
Technical Practice Under Scrutiny
Distillation is a widely used machine learning technique in which a smaller "student" model is trained on outputs from a more advanced "teacher" model, reducing computational costs. The technique's legal status remains ambiguous, and US firms have used similar methods among themselves (Source: SCMP).
However, US policymakers are increasingly framing the practice as a national security threat rather than a technical optimization. Anthropic has specifically accused three AI laboratories—DeepSeek, Moonshot, and MiniMax—of attempting to copy its models through coordinated distillation campaigns involving synchronized accounts and "jailbreak" attempts designed to extract proprietary information (Source: BBC).
Chinese Companies Named, Beijing Denies Allegations
The administration's focus has fallen particularly on DeepSeek, whose low-cost AI model attracted global attention last year. DeepSeek launched a preview of its V4 model on Friday, adapted for Huawei chip technology—a development analysts say demonstrates growing Chinese autonomy in the sector (Source: Dawn).
The Chinese Embassy in Washington rejected the accusations as "groundless" and "deliberate attacks on China's development and progress in the AI industry." A representative stated: "China is not only the world's factory but is also becoming the world's innovation lab. China's development is the result of its own dedication and effort as well as international cooperation" (Source: Dawn; Source: Guardian).
DeepSeek has previously stated that its V3 model used "naturally occurring data collected through web crawling" and had not intentionally used synthetic data generated by OpenAI (Source: Dawn).
Competitive Context: 'Effectively Closed' Gap
The heightened tensions come as experts assess that China's AI capabilities have advanced considerably. A report from Stanford University's Institute for Human-Centered AI concluded that the US-China gap in performance of top AI models has "effectively closed" (Source: Guardian). The US had maintained an early lead with models like OpenAI's ChatGPT, but Chinese developments throughout 2024-2025 have significantly narrowed that advantage.
Some analysts predict that the administration's threatened crackdown could force consolidation in China's AI sector. Zhang Ruiwang, a Beijing-based information systems architect, suggested that Chinese firms lacking original research could be "forced out of the game" within six to 12 months, while development cycles for more capable developers could lengthen from three months to a year or more (Source: SCMP).
Policy Paradigm Shift
The controversy reflects what sources describe as a fundamental shift in Washington's approach to AI governance. Where US policy once focused primarily on AI safety, ethical risks, and algorithmic harms, it has moved toward a "security-driven paradigm" centered on strategic competition and technological control (Source: SCMP). This includes Washington's push for what has been described as a "full-stack AI export" system to secure supply chains and limit Chinese access to advanced chips.
The timing of these developments coincides with expectations of a Trump visit to China in May, adding a bilateral diplomatic dimension to the escalating technology competition (Source: BBC; Source: Dawn).
Visa and Immigration Scrutiny
Separate from the technology-specific measures, congressional hearings have also examined the US visa system as a potential vulnerability. Senator Ashley Moody of Florida told the Senate Judiciary Committee that the US was rolling out the "red carpet" for spies through permissive legal channels, expressing concern about nations with laws requiring citizens to gather intelligence (Source: SCMP). These allegations were not specifically substantiated in the source material available.