The White House on Thursday accused foreign actors—primarily in China—of mounting “industrial‑scale” campaigns to copy the capabilities of U.S. AI models, using tactics like jailbreaking and vast networks of fake accounts to extract proprietary outputs and train competing systems.
In a memorandum titled “Adversarial Distillation of American AI Models,” Michael Kratsios, Assistant to the President for Science and Technology and Director of the Office of Science and Technology Policy (OSTP), said the U.S. government has evidence of coordinated distillation efforts aimed at U.S. frontier AI systems. Kratsios posted on X that “the U.S. has evidence that foreign entities, primarily in China, are running industrial‑scale distillation campaigns to steal American AI. We will be taking action to protect American innovation.”
What the government says is happening
- Attackers are reportedly using “tens of thousands of proxy accounts” and jailbreak techniques to systematically extract model outputs — a pattern the memo calls a distillation campaign.
- A distillation attack trains a smaller model by feeding it the outputs of a larger, protected model. The resulting copy may not fully match the original but can mimic performance on many benchmarks at far lower cost.
- The administration warned that unauthorized distillation can also strip away safety and content controls meant to keep systems “ideologically neutral and truth‑seeking.”
A real‑world example
In February, Anthropic accused Chinese AI labs DeepSeek, Moonshot, and MiniMax of extracting millions of responses from Claude using roughly 24,000 fraudulent accounts to train rival models — an incident consistent with the distillation tactics described in the OSTP memo.
U.S. response and next steps
The memo says federal agencies will:
- Coordinate with U.S. AI companies to strengthen protections around frontier models,
- Work with private industry to develop defenses against large‑scale distillation campaigns, and
- Explore mechanisms to hold foreign actors accountable for unauthorized copying.
OSTP also acknowledged that lawful distillation can produce useful smaller, open‑weight models, but drew a line at “systematically extracting and copying the innovations of American industry.” The office did not immediately respond to a request for comment by Decrypt.
Why crypto watchers should care
AI models increasingly power trading algorithms, on‑chain data analysis, smart-contract auditing tools and market intelligence used across the crypto ecosystem. Large‑scale theft or degradation of safeguards could accelerate the spread of lower‑cost but less safe models into tooling and services that crypto projects rely on — raising risks for security, manipulation, and misinformation.
Bottom line: Washington says it has evidence of organized, large‑scale efforts to siphon U.S. AI capabilities and is mobilizing government and industry defenses. The development bears watching for both AI and crypto communities as enforcement and technical countermeasures evolve.
Read more AI-generated news on: undefined/news
