Alignment remains the central strategy for improving AI capability while restricting dangerous knowledge.

In the last six weeks, Anthropic’s most protected model, Fable 5, was temporarily shut down after a letter from the surveyed Trump administration and was soon overtaken on leader‑boards by a Chinese model. The Chinese model, Kimi K3 from Moonshot AI, was launched in mid‑July, and its weights—trillions of numbers that encode everything the model knows—will be publicly released on July 27.

The White House accuses Moonshot of using responses from Fable 5 to train Kimi K3 in a process known as model distillation. If the claim is accurate, the shut‑down model lives on inside its Chinese rival. The allegation raises questions about cross‑border AI training and the integrity of proprietary data.

Treasury Secretary Scott Bessent has warned that sanctions and trade‑limiting entity‑list designations could be the next step. The situation underscores the urgency of clear regulatory frameworks to keep AI development safe and transparent.