The Trump administration leveled explicit accusations Thursday that China-based actors are systematically stealing American artificial intelligence technology through what a top White House official called "deliberate, industrial-scale campaigns."
Michael Kratsios, director of the White House Office of Science and Technology Policy, sent a memo to federal agency heads describing coordinated efforts to extract proprietary information from leading U.S. AI models including Claude and Gemini. The actors use proxy accounts to hide their tracks and employ jailbreak techniques designed to unlock restricted capabilities, according to the memo.
The theft method relies on a technique known as distillation attacks. Operatives repeatedly query proprietary models through API access, executing millions of queries to build datasets that allow them to replicate how the systems function. The goal is to create knockoff versions that perform at comparable levels to American models but cost far less to develop.
Kratsios warned that these tactics accomplish more than just copying AI performance. They also strip away safety guardrails designed to keep the systems producing ideologically neutral and factually accurate output.
Earlier this year, both OpenAI and Anthropic identified Chinese firms behind large-scale distillation campaigns, naming DeepSeek, Moonshot AI, and MiniMax as primary actors. The allegations arrive as Trump prepares for a significant diplomatic visit to Beijing next month where he is expected to negotiate economic concessions and reshape elements of the U.S.-China relationship.
The White House plans to share intelligence about these campaigns with American AI companies, providing details on the tactics being deployed and helping the private sector develop countermeasures.
Kratsios expressed skepticism about the long-term viability of models built through these stolen foundations. "As methods to detect and mitigate industrial-scale distillation grow more sophisticated, foreign entities who build their AI capabilities on such fragile foundations should have little confidence in their integrity and reliability of the models they produce," he wrote.
The accusation carries significant weight within the broader context of U.S.-China technology competition. The Justice Department indicted a former Google software engineer in 2024 for stealing AI trade secrets and providing them to two Chinese companies, part of a longstanding pattern of intellectual property theft allegations against China.
Author James Rodriguez: "This is a shot across the bow, but it's worth noting the timing creates real tension just as Trump heads into delicate negotiations with Beijing."
Comments