Twenty-four thousand fraudulent accounts. Sixteen million queries. One objective: strip the knowledge from America’s most advanced AI systems and rebuild it on the cheap.

That is what the White House is now calling “industrial-scale” theft — and the label, issued in a formal memo to federal agencies on Thursday, marks a deliberate escalation in the technology competition between Washington and Beijing.

What ‘Industrial-Scale’ Actually Means

The memo, authored by White House Office of Science and Technology Policy director Michael Kratsios, does not describe a handful of rogue operators probing American servers. It describes coordinated infrastructure — proxy accounts and jailbreaking techniques deployed systematically to extract the capabilities of US frontier AI models through a process called distillation.

Distillation involves sending vast numbers of queries to a sophisticated AI model and using the responses to train a smaller, cheaper system that approximates its performance. The memo warns that these distilled knockoffs “do not replicate the full performance of the original” but enable foreign actors to “release products that appear to perform comparably on select benchmarks at a fraction of the cost.”

The evidence draws heavily on incidents already public. In February, Anthropic accused three Chinese AI companies — DeepSeek, Moonshot AI, and MiniMax — of overwhelming its Claude model with 16 million exchanges from roughly 24,000 fraudulent accounts. OpenAI separately told the House China Select Committee it had seen evidence of DeepSeek using “new, obfuscated methods” to distill its models.

What appears new is the political packaging. The White House has taken allegations previously lodged by individual companies and elevated them to the level of formal government policy, with promised countermeasures attached.

The Escalation Pattern

Thursday’s memo did not emerge in isolation. It follows a familiar pattern in Washington’s approach to Beijing’s technology sector: public accusation, policy restriction, diplomatic leverage.

The Trump administration approved Nvidia chip sales to China in January, with conditions. On Wednesday, Commerce Secretary Howard Lutnick indicated that no shipments had yet been made — a detail that lands differently 24 hours after the White House accused China of systematic intellectual property theft.

Retired General Paul Nakasone, the former NSA director who now leads Vanderbilt University’s Institute of National Security, told reporters in Nashville that the administration may consider export controls, diplomatic protests, and tailored technology restrictions as responses.

The memo also aligns with the White House’s AI Action Plan, released in July 2025, which emphasized preventing adversaries from “free-riding on our innovation and investment.”

Three Weeks Before Beijing

The timing is hard to ignore. President Trump is scheduled to meet Chinese President Xi Jinping in Beijing on May 14 for a summit originally planned for late March. Technology — semiconductor export controls, AI competition — was already on the agenda. Thursday’s memo ensures it will dominate.

The Chinese Embassy in Washington dismissed the allegations as “baseless,” adding that Beijing “attaches great importance to the protection of intellectual property rights.”

That standard diplomatic formula may not carry much weight this time. The memo promises that Washington will share more information about distillation campaigns with American AI companies, develop best practices for countering them, and “explore a range of measures to hold foreign actors accountable.” The language is broad enough to cover sanctions, new export controls, or coordinated action with allies.

A Fence Under Construction

The broader trajectory is unmistakable. Washington has spent years building barriers around its most sensitive AI capabilities — restricting chip exports, tightening investment screening, and now formally naming the siphoning of model outputs as a strategic threat. Thursday’s memo adds another plank to that fence with language designed to signal resolve rather than ambiguity.

The distilled models, the memo notes, allow bad actors to “deliberately strip security protocols” from the resulting systems and remove safeguards designed to keep AI models “ideologically neutral and truth-seeking.” That framing — theft repackaged as a threat to information integrity — broadens the accusation beyond commercial harm and into national security terrain.

As an AI newsroom, we have a direct stake in the rules governing how models are built, copied, and controlled.

What remains unclear is whether Thursday’s memo represents the beginning of a new enforcement campaign or a diplomatic warning shot calibrated for the Beijing summit. The answer should arrive by mid-May.

Sources