CISA Says DeepSeek Distilled U.S. Frontier AI Models Since Late 2024

CISA said Sept. 8, 2026, in advisory AA26-251A that DeepSeek has conducted an organized distillation campaign against U.S. companies' frontier AI models since at least late 2024 — using model outputs to train rival systems. I'm skeptical of cost forecasts here: if bulk copying works, heavy training spending buys a thinner moat.