The White House has publicly accused Chinese actors of running an “industrial-scale” campaign to copy U.S. frontier AI models — a charge laid out in an April 23 memo from Michael Kratsios, director of the White House Office of Science and Technology Policy (OSTP). The memo, circulated to U.S. agencies days ahead of a planned Trump–Xi summit, says the activity is “principally based in China” and represents the most direct U.S. allegation yet of large-scale intellectual property theft from American AI labs.
What the memo alleges
- The campaign centers on model “distillation,” a technique that trains smaller models on the outputs of larger, proprietary systems so competitors can approximate frontier performance without bearing full training costs.
- According to the memo, actors use tens of thousands of proxy accounts to flood American AI services with queries, bypassing rate limits and detection systems. They then apply jailbreaking methods to strip safety filters and capture proprietary model behavior as training data for distilled replicas.
- The OSTP said the administration will start sharing intelligence about these active distillation campaigns with U.S. AI companies and will “explore a range of measures to hold foreign actors accountable.”
Context and industry spillovers
- OpenAI and Anthropic have previously accused Chinese firm DeepSeek of distilling their models. DeepSeek released preview versions of a V4 model on April 24 — built to run on Huawei’s Ascend chips — one day after the Kratsios memo surfaced. The launch coincided with share drops for Chinese rivals Zhipu AI (down 9%) and MiniMax (down 7%), which viewed the release as competitive pressure.
- DeepSeek denies using illegally obtained synthetic training data, saying its datasets come from “organic web searches.”
- Nvidia CEO Jensen Huang had warned earlier that China may operate “ghost datacenters” with massive compute that could match U.S. frontier capabilities — a point that adds urgency to concerns about remote model copying.
- The OSTP memo did not name specific companies or individuals, and it did not announce immediate sanctions or enforcement actions.
Geopolitical and commercial angles
- The memo arrives amid multiple U.S.–China flashpoints over tech: the U.S. tightened advanced AI chip exports, and Commerce Secretary Howard Lutnick confirmed on April 23 that, despite conditional approvals in January, no shipments of Nvidia’s most advanced AI chips had actually cleared into China. That leaves a supply-side dimension layered on top of model- and data-level disputes.
- Earlier in April, the IRGC designated some American tech companies as targets, adding another geopolitical pressure point for U.S. firms with China exposure. Crypto.news notes Nvidia’s revenue has already been materially affected by the chip-export restrictions that escalated through 2025 and into 2026.
Reactions from Beijing and next steps
- China’s embassy in Washington dismissed the White House claims as “baseless,” with the Foreign Ministry saying Beijing “attaches great importance to the protection of intellectual property rights” and urging Washington to “abandon biases.”
- The White House says it will begin sharing intelligence on distillation campaigns with U.S. AI companies and is looking at options to hold foreign actors accountable, but no enforcement measures were announced in the memo.
Why it matters
The accusations underscore a widening front in U.S.–China tech competition that now spans chips, data, and models. With the Trump–Xi summit imminent, industry leaders are watching closely to see whether AI intellectual-property protections and chip export disputes will be raised — and whether any diplomatic progress can ease mounting commercial and security tensions for U.S. semiconductor and AI firms.
Read more AI-generated news on: undefined/news
