The White House has privately accused China of "industrial-scale" theft of intellectual property from U.S. artificial-intelligence labs, according to reporting that surfaced Thursday. At the center of it all is a government memo that reportedly flags coordinated efforts to extract advanced model capabilities at scale from American systems. If confirmed, the claim would mark a sharp escalation in the tech rivalry between U.S. companies and Chinese developers.
As relayed by Reuters, the Financial Times reported that the memo was written by Michael Kratsios and warned of "industrial-scale distillation" campaigns that rely on tens of thousands of proxy accounts and jailbreaking techniques to harvest proprietary outputs. According to that reporting, the memo instructed agencies to share technical indicators with American AI firms and said the administration would explore measures to hold foreign actors accountable. Reuters added that it could not immediately verify the FT report and that officials had not commented when contacted.
What the Memo Flags
The memo, according to those reports, describes patterns U.S. officials view as distinct from ordinary API use: synchronized, repetitive queries, coordinated networks of accounts, and prompts engineered to expose internal reasoning traces. Officials say those signals point to deliberate "distillation" campaigns intended to capture model behavior without building the underlying training pipelines. Michael Kratsios's role as director of the White House Office of Science and Technology Policy is public and listed on the White House website.
Industry Warnings and Evidence
The memo's claims track closely with warnings U.S. firms have been issuing since February. Anthropic published a technical disclosure on Feb. 23 alleging three Chinese labs — DeepSeek, Moonshot and MiniMax — used roughly 24,000 fraudulent accounts and more than 16 million exchanges to "distill" its Claude model. OpenAI also sent a memo to the House in February describing similar distillation tactics and advising lawmakers of techniques companies are seeing in the field, with the OpenAI memo publicly available in committee filings.
Enforcement and the Chip Backdrop
Those model-theft allegations land as hardware controls and enforcement remain central to policy debates. U.S. Commerce officials have said no shipments of Nvidia's H200 chips to China have been completed so far, underscoring tight export controls, according to Reuters. At the same time, federal prosecutors in March charged three people tied to a server maker in an alleged $2.5 billion diversion of servers to China, a case reported by Bloomberg. Together, the hardware and model claims highlight how difficult it is for regulators and companies to police both the movement of advanced chips and the large-scale misuse of model outputs.
Why Silicon Valley Labs Should Pay Attention
Frontier AI work is concentrated in the Bay Area, and local labs are central actors in both the technical contest and the policy response. If distillation at scale becomes common, companies may tighten API access, accelerate detection tooling, or increase intelligence-sharing with agencies to protect trade secrets and safety guardrails. Any of those shifts would change how researchers, startups and enterprise customers in Silicon Valley interact with advanced models and with one another.
For now, the story is still unfolding. Watch for any White House briefings, further reporting from the Financial Times and Reuters, and follow-on enforcement filings that could make the memo's allegations public. The coming weeks will show whether the administration moves from warning to formal action, and how the industry adapts to the prospect of coordinated capability extraction.
