White House accuses China of copying American AI models in ‘industrial-scale’ campaign | CNN Business - CNN

Ah, just what the AI cold war needed: a memo. White House director Michael Kratsios has officially announced that China is conducting 'industrial-scale' campaigns to distill US frontier AI models — meaning they're running inference on American models to cheaply steal their general capabilities by labelling synthetic data. Never mind that the best use of this intelligence, according to the memo, is for supply chain management and manufacturing robot design, which sounds less like 'existential threat' and more like 'Chinese warehouses just got 12% faster at packing boxes.' The US response? Convene 'other intelligence-led governments' to presumably submit a strongly worded letter. This follows the joyful news that DeepSeek had already released V4 — a massive distraction campaign now canonised as a security breach. One has to admire the audacity: Washington wants to keeps models locked up to defend against distillation, yet companies like OpenAI have been serving up these same outputs in a ChatGPT interface for years, practically marinated in prompt-injection steroids. The great irony is that the very thing the White House calls a 'cost-based safety conflict' — not just copying but actual safety concerns — means the deeper worry here isn't theft of capabilities. It's that Chinese models might replicate the American models' tendency to produce wrong, bigoted, or nanny-ish answers at scale.