White House Alleges ‘Industrial-Scale’ AI Theft by China, Signaling Enforcement Pivot

The Trump administration formally accused Chinese entities of conducting ‘industrial-scale’ campaigns to extract proprietary capabilities from U.S. frontier AI systems, marking a strategic shift from hardware export controls to direct intellectual property enforcement. White House Office of Science and Technology Policy Director Michael Kratsios stated in an April 23 memo that foreign entities, principally based in China, are engaged in deliberate distillation attacks leveraging tens of thousands of proxy accounts and jailbreaking techniques to reproduce the core capabilities of American models, according to Invezz.

The allegation arrives as the performance gap between leading U.S. and Chinese AI models has collapsed to just 2.7%, despite American firms outspending Chinese competitors 23-to-1 on development. China’s Dola-Seed-2.0 now trails Claude Opus 4.6 by a narrow margin despite development costs of $12.4 billion versus $285.9 billion for U.S. systems, per the Stanford AI Index 2026. That convergence is occurring even as U.S. Export Controls on advanced chips remain in effect — suggesting distillation has become China’s primary pathway to frontier capability.

The Distillation Campaign

Anthropic disclosed in February that DeepSeek, Moonshot AI, and MiniMax generated over 16 million conversations with Claude using more than 24,000 fake accounts, according to Euronews. The coordinated nature of the attacks — distributed across proxy infrastructure to evade detection — indicates systematic extraction rather than isolated research activity. OpenAI, Anthropic, and Alphabet announced plans on April 6 to share threat intelligence through the Frontier Model Forum in response to the campaign, per the Foundation for Defense of Democracies.

The Kratsios memo cited analyst Chris McGuire’s assessment that Chinese AI firms are relying on distillation attacks to offset deficits in computing power and illicitly reproduce core capabilities. The allegation reframes distillation from research methodology to industrial espionage — a critical legal distinction that enables prosecution under the Economic Espionage Act. Former Google engineer Linwei Ding was convicted on January 30 on seven counts of economic espionage and seven counts of trade secret theft for stealing AI technology including TPU and GPU designs, demonstrating the Justice Department’s willingness to pursue individual prosecutions, per The Register.

Policy Response Incoming

The House Foreign Affairs Committee passed legislation requiring consideration of placing Chinese companies involved in distillation attacks on the entity list — a designation that would block access to U.S. technology and financial systems. The Senate Judiciary Committee held a hearing titled ‘Stealth Stealing: China’s Ongoing Theft of U.S. Innovation’ on April 22, one day before the Kratsios memo, signaling coordinated pressure across branches, according to IP Watchdog.

The enforcement pivot represents a recalibration after the Trump administration’s January 15 decision to allow export of NVIDIA H200 chips to China under case-by-case licensing proved strategically incoherent. Chinese AI firms placed orders for more than 2 million H200 chips worth up to $14 billion for 2026, undermining the original intent of chip export controls, per Bloomsbury Intelligence. Applied Materials was fined $252 million on February 12 for illegally exporting ion implantation equipment to China — the second-largest penalty in Bureau of Industry and Security history — demonstrating that hardware controls are difficult to enforce and frequently violated.

China’s Structural Advantages Persist

Even as the U.S. pivots to IP enforcement, China continues consolidating infrastructure advantages that enable efficient model development. China added over 540 gigawatts of new power capacity in 2025 — about 80% solar and wind — equivalent to building the entire U.S. power grid over the past four years, according to testimony before the Brookings Institution. That energy surplus allows Chinese labs to train models at lower marginal cost than U.S. competitors facing grid constraints and higher electricity prices.

China leads in AI patents with 69.7% of global filings and industrial robot installations at nine times the U.S. rate. AI talent migration to the United States has dropped 89% since 2017, with 80% of that decline occurring in the last year alone — a reversal driven by visa uncertainty, research collaboration restrictions, and competitive salaries in Chinese labs.

What to Watch

The Kratsios memo sets the stage for entity list designations targeting Chinese AI firms involved in distillation — likely announced before Trump’s planned Beijing summit. Expect tightened API access controls from U.S. labs, mandatory third-party audits of high-volume users, and prosecution of intermediaries facilitating proxy account networks. The House Foreign Affairs Committee’s AI Overwatch Act, passed January 22, gives Congress 30-day review authority over AI chip export licenses — a mechanism that could be repurposed to scrutinise model access and collaboration agreements.

Monitor whether enforcement extends beyond model distillation to university research partnerships. Chinese institutions account for 23.2% of global AI publications, much of it co-authored with U.S. labs — a collaboration network now under scrutiny as potential IP leakage pathway. If enforcement targets academic partnerships, expect talent flight to accelerate and U.S. dominance in AI research to erode further. The 2.7% performance gap is narrow enough that a single breakthrough — whether stolen or indigenous — could flip leadership. The U.S. bet is that IP enforcement will slow China’s convergence. The risk is that structural advantages in energy, cost efficiency, and talent retention render enforcement irrelevant.