White House ties China to industrial-scale distillation attacks
Original: White House accuses China of 'deliberate, industrial-scale campaigns' to steal US AI models View original →
Washington is no longer treating distillation as a gray-area abuse case
The sharpest change in the White House memo is categorical. Model distillation is being framed as theft of strategic technology, not as a platform-misuse problem for AI labs to handle alone. Nextgov reports that the Office of Science and Technology Policy accused China and other foreign entities of running deliberate, industrial-scale campaigns against U.S. frontier systems. If that framing holds, the policy response moves from account bans and rate limits toward export controls, diplomatic pressure, and broader national-security coordination.
Why the memo cares about benchmark parity
The memo's most revealing line is not that copied models fall short of the original. It is that they can still look competitive where it counts in public comparisons. The document says models built from unauthorized distillation may fail to reproduce the full performance of the source system, yet can appear comparable on select benchmarks at a fraction of the cost. That is precisely the scenario U.S. labs fear: an attacker does not need perfect theft if a cheaper imitation is good enough to win attention, customers, or geopolitical leverage.
The backdrop is no longer hypothetical
Nextgov ties the memo to several recent claims from frontier labs. Anthropic said in February that DeepSeek, Moonshot AI, and MiniMax had flooded Claude with 16 million exchanges from roughly 24,000 fraudulent accounts. OpenAI separately told members of the House China Select Committee that it had seen evidence consistent with ongoing attempts by DeepSeek to distill frontier models through obfuscated methods. The memo itself reportedly stops short of naming companies, but OSTP Director Michael Kratsios said the campaigns use tens of thousands of proxies and jailbreaking techniques to systematically extract American breakthroughs.
What changes from here
The practical policy shift is in the follow-through. OSTP told agencies the administration will expand information-sharing with the private sector, work with companies on defensive best practices, and look for new ways to hold foreign actors accountable. Retired Gen. Paul Nakasone told reporters that export controls, diplomatic protests, and tailored technology restrictions could all become part of the toolkit. For companies building models or depending on frontier APIs, this matters because the fight over distillation is moving out of the trust-and-safety silo and into trade and national-security policy. The Nextgov report and the linked White House memo are the documents to watch.
Related Articles
The policy fight is no longer just about model benchmarks. Axios reports that U.S. officials have revisited tools such as Entity List threats, security advisories, procurement pressure, and hosting liability rules as cheaper Chinese open-weight models gain enterprise traction.
Sen. Mark Warner’s July 21 AI agenda ties mandatory secure testing for advanced models to agent rules, data-center disclosure, and workforce-transition funding. The package shows U.S. AI policy expanding from narrow model safety into infrastructure, markets, labor, and national security.
Build American AI, a nonprofit linked to a $100M super PAC backed by OpenAI and Palantir executives, is running a campaign paying influencers up to $5,000 per TikTok video to frame China's AI advancement as a threat to American safety and jobs. Exposed by Wired, the two-phase campaign promotes US AI positively while stoking fear of Chinese AI.