The window is closing fast. Anthropic released a stark warning Thursday: the United States has between 12 and 24 months to lock in a decisive lead in artificial intelligence before China catches up. Miss that window, and the advantage evaporates.

The AI company laid out two futures for 2028. In one scenario, the US restricts China's access to American computing power and closes policy loopholes. In the other, it doesn't.

The difference between those two worlds, Anthropic argues, is a decisive technological edge or a "neck-and-neck race" that neither country can afford.

China isn't waiting passively. Anthropic identified two primary tactics draining the US advantage. First, Chinese companies exploit loose controls on chip exports.

Nvidia and AMD chips still find their way to Chinese labs through smuggling operations; a December bust revealed underground networks relabelling advanced Nvidia GPUs as generic "SANDKYAN" chips.

Second, Chinese AI firms use distillation attacks, training smaller "student" models using Claude and other advanced American AI systems.

In February, Anthropic named names, including DeepSeek, MiniMax, and Moonshot AI, that were illicitly using Claude to accelerate their own development. The technique is elegant and devastating; it lets Chinese labs leapfrog years of research without needing their own compute resources.

Anthropic wants to strengthen chip export controls, increase enforcement budgets, and implement measures to stop distillation attacks. The company framed it bluntly, stating, "Our past success means that our present task is largely to avoid squandering our advantage."