The White House on Thursday publicly accused China of running 'industrial-scale' campaigns to siphon capabilities from the most advanced US artificial intelligence systems, in a memorandum from the Office of Science and Technology Policy that signals a sharp escalation in Washington's technology posture three weeks before President Trump is scheduled to meet Xi Jinping in Beijing.
The memo, authored by OSTP director Michael Kratsios, alleges that foreign entities are deploying 'tens of thousands of proxy accounts' alongside jailbreaking techniques to 'systematically extract capabilities from American AI models,' according to reporting by the Financial Times and subsequent summaries from Fox Business, U.S. News and GV Wire. Kratsios framed the activity as coordinated rather than opportunistic and warned that 'we will be taking action to protect American innovation.'
How the alleged theft works
At the center of the accusation is a technique known as model distillation. A smaller model is trained on the outputs of a larger, more capable one — in effect cloning the teacher's behavior for a fraction of the training cost. The White House argues that by fanning out across frontier APIs through huge pools of proxy accounts, adversaries can evade rate limits and detection, then strip security protocols from the derivative models before redeploying them.
The memo describes the resulting copies as brittle. 'Foreign entities who build on such fragile foundations should have little confidence in the integrity and reliability of the models they produce,' Kratsios wrote, in a line that appears aimed both at Chinese labs and at their downstream customers. Reporting cited in the coverage pointed to Anthropic's Claude and references to DeepSeek as part of the backdrop, though the memo stops short of naming specific Chinese labs as defendants.
What Washington says it will do
The administration committed to two concrete steps: sharing 'distillation threat' information with American AI companies, and exploring 'a range of measures to hold foreign actors accountable.' Specific enforcement tools — export controls, sanctions, or API-level identity requirements — were not detailed in the document.
The Chinese Embassy in Washington declined to comment immediately, according to GV Wire.
Summit context
The timing is conspicuous. Trump is scheduled to sit down with Xi on May 14 in Beijing, a meeting expected to cover semiconductor export licenses and AI policy. The memo lands shortly after January's approval for limited Nvidia chip shipments to China — shipments that Commerce Secretary Howard Lutnick said this week had not yet occurred.
Implications for AI developers
For frontier labs, the memo is effectively an invitation to tighten identity verification on high-capability APIs, which several vendors have already moved toward. It also gives political cover for stricter gating of open-weight releases and may accelerate the industry-wide push toward verified-developer programs. For enterprise buyers, the message is quieter but real: models built on uncertain provenance now carry reputational as well as security risk.
