The Essay That Started With a Specific Year
Anthropic published an essay. The title: "2028: Two Scenarios for Global AI Leadership."
Not 2030. Not "the coming decade." 2028. Two years from now. They chose that date deliberately and the body of the essay explains why: the decisions being made right now , about export controls, compute access, research talent, and policy , will have locked in the outcome by then. The window to influence which scenario plays out is open now and will not be open in three years.
The reviewer covering this essay led with a single sentence: "Anthropic is panicking."
The Two Scenarios
Scenario one: The United States and its democratic allies maintain their lead in AI development through 2028 and beyond. They set the terms for how the technology is developed and deployed. The norms that govern the most powerful AI systems in history are shaped by democracies.
Scenario two: Authoritarian governments , the essay names China's Communist Party specifically , close the gap or pull ahead. They set the terms. The norms are theirs.
Anthropic's argument is not that China is evil and America is good. It is that the political system in which the most advanced AI is developed will shape how that AI is used. AI gives a government the ability to repress citizens at unprecedented scale , not by hiring more enforcers, but by automating surveillance, analysis, and response in ways that no human government has ever been able to deploy. Under a democratic system, there are constraints on those uses. Under an authoritarian system, there are not.
"AI under an authoritarian regime," the essay states, "can be more harmful than anything we have seen previously in history."
Why China Is Right Behind Us
The conventional assumption is that US export controls on chips are keeping China behind. The essay complicates this.
China is not close to the frontier because of distillation attacks on Western models. They are close because of raw algorithmic innovation and raw human talent. The export controls slow the hardware gap. They do not eliminate the talent and research gap, which is smaller than most people assume.
The essay's conclusion is that the current lead is real but not permanent. It requires active policy work , export controls, anti-distillation measures, talent retention, and international coordination with democratic allies , to maintain. Doing nothing while assuming the lead will hold is the scenario that ends badly.
The Thing About Whoever Is in Control
The geopolitical argument gets abstract quickly. Here is the concrete version.
Whoever leads in AI gets to decide what "safe AI" means. What it means for a model to refuse a request. What information systems are allowed to surface. What counts as harmful output and what counts as acceptable. These are not technical questions , they are political ones, answered by the political system that built the technology.
If the answer to "who decides?" is the United States and its allies, the answer to "what gets decided?" looks one way. If the answer is a government that surveils its citizens, prosecutes dissent, and treats international norms as obstacles, it looks another way.
Anthropic published this essay publicly. They are not a government body. They are a private AI company with significant commercial interests in the outcome. Read it with that in mind.
Read it anyway. The 2028 deadline is not a prediction. It is a warning with a specific expiration date attached , and the date is close enough to matter.